Method and device for background segmentation on basis of movement localization

FIELD: movement detection systems, technical cybernetics, in particular, system and method for detecting static background in video series of images with moving objects of image foreground.

SUBSTANCE: method contains localization of moving objects in each frame and learning of background model with utilization of image remainder.

EFFECT: increased speed and reliability of background extraction from frames, with possible processing of random background changes and camera movements.

4 cl, 14 dwg

 

The technical field

This invention relates to the field of motion detectors or motion detection systems and, in particular, relates to a method and device for segmenting the background on the basis of the localization movement.

The level of technology

Video conferencing and automatic surveillance are rapidly developing fields of technology, growth aided by the increasing availability of systems with low cost and progress in the field of technology of motion detection. Technique video provides sequentially displaying images using the device, which displays images, such as a computer display. The sequence of images varies in time, so that it can adequately represent the movement in the scene.

The frame is a single image in an image sequence, which is transmitted to the monitor for display. Each frame consists of picture elements (absorbing or pixels), which are the basic unit of programmable color in the image or frame. A pixel is the smallest area of the monitor screen, which can be turned on and off to create the image, while the physical size of a pixel depends on the resolution of the computer display. The pixels may be formed in rows and columns of the display is the computer to render the frame. If the frame contains a color image, each pixel can be enabled with a specific color to render the frame. Specific color, which gives a pixel is a mixture of components of the color spectrum, typically such as red, green, and blue.

Video sequences can contain as stationary objects and moving objects. Stationary objects are objects that remain stationary from one frame to another. Thus, the pixels used to render colors a stationary object, remain essentially the same in consecutive frames. Zone of the frame containing objects with the same color, called background. Moving objects are objects that change position in the frame relative to the previous position within the previous frame in the sequence of images. If the object changes its position in the next frame relative to its position in the previous frame, the pixels used to render images of the object change color in consecutive frames. Areas of the frame are called the foreground.

Some applications, such as the technique of video, often based on the detection of the motion of objects in video sequences. In many systems, this motion detection is based on proofread and background. Background subtraction is a simple and effective method for the identification of objects and events of interest in video sequences. A substantial stage of background subtraction is a model training background the ability to explore a private environment. In most cases, this implies obtaining background images for later comparison with the test image that can be foreground objects. However, this approach faces problems in applications where the background is not present or is changing rapidly.

Some methods according to the prior art, which are aimed at solving these problems, often referred to as segmentation of the background. Approaches to the problem of segmentation of the background can be roughly divided into two stages: segmentation of movement and training background. The motion segmentation is used to locate each frame of the sequence of zones that correspond to moving objects. Segmentation of motion starting from the field of motion derived from the optical flow calculated in two consecutive frames. Field movements are divided into two clusters using k-means. The largest group is taken for the background.

Training background is in training the model background on the rest of the image. A model-based background subtraction is to subtract the background from the "Museum" color images on the new assumptions about the properties of the image. This includes a small number of objects in the background, which is relatively smooth with color changes in space and light texture.

The disadvantage of these solutions segmentation of background, according to the prior art, is that they offer a pixel based approach to the segmentation of motion. When pixel based approach analyzes each pixel for making decisions, whether it is to the background or not. Therefore, the time T of the processing of each pixel is the sum of the time T1, the motion detection and time T2 training background. If a frame consists of N pixels, the processing time per frame is T*N. This approach can be quite reliable, but requires a lot of time.

Brief description of drawings

Below is the description of the invention using a specific example with reference to the accompanying figures of the drawings, which do not limit the invention, with the figures of the drawings shows the following:

figure 1 - depicts an embodiment of the method of selecting the background image from the video sequence;

figa is an example of a frame from the video sequence;

figv is another example of a frame from the video sequence on the next frame, according figa;

figs - an example of performing image change detection;

fig.2D sample execution paths Gras the Itza change detection image, according figs;

file is an example of the construction of the shell;

figure 3 - the option to perform an iterative design environment;

4 is an embodiment of circuit training background;

figure 5 - example of the relative variation of the current average values depending on a;

6 is an example of the signs of tracking example of the background frame;

7 is an embodiment of a motion detection camera and its compensation;

Fig is an example of the percentage of the number of moving pixels of the segmented using the algorithm of localization movement;

Fig.9 is an example of the percentage of the number of background pixels that are segmented as foreground, obtained using the algorithm of localization movement;

figure 10 - example of a complete computer system with camera.

Detailed description

In the following description numerous specific details are described as examples of specific systems, technologies, components, etc. to ensure a deep understanding of this invention. However, for professionals in the art it is obvious that these specific details are not required to implement the invention. In other instances, well-known from the prior art components and methods are not described in detail to avoid unnecessarily complicating the description.

This izobreteny which contains different stages outlined below. Stage of the present invention can be performed using components of hardware or may be presented in a running machine commands that can lead to the execution of the General-purpose processor or a special processor, programmable commands that are specified stages. As an alternative solution, the stage can be performed using a combination of hardware and software.

This invention can be offered in the form of a computer program or software, which may include a machine-readable storage medium having stored therein the commands that can be used to program a computer system (or other electronic devices) to perform the method according to the invention. Machine-readable media includes any mechanism for storing or transmitting information in a form (e.g. software, application processing, and so on), readable by a machine (e.g. computer). Machine-readable media may include, but are not limited to, magnetic storage medium (e.g. floppy disk), optical recording media (e.g. CD-ROM), magneto-optical recording medium; p is the permanent memory (ROM), memory (RAM), erasable programmable memory (e.g., EPROM and EEPROM), flash memory, electrical, optical, acoustic signal or a signal with a different form of distribution (e.g., carrier waves, infrared signals, digital signals, etc), or other types of recording media suitable for storing electronic commands.

This invention can also be implemented in a distributed computing system, where the machine-readable storage medium is stored and/or executed by more than one computer system. In addition, information transmitted between computer systems, can be spread by polling changes or method forced distribution in the communication environment, connecting computer systems.

Some parts of the descriptions presented in the form of algorithms or symbolic representations of operations on data bits that can be stored in memory and which can operate the computer. These algorithms and representations are the means used by experts in the field of technology for the efficient performance of their work. The algorithm is usually understood as a self-contained sequence of actions leading to the desired result. Actions are actions that require treatment with quantities. Usually, but not necessarily, these Koli is estva take the form of electrical or magnetic signals, which can be stored, transferred, combined, compared, etc. it is Sometimes convenient for General use to refer to these signals, bits, values, elements, symbols, characters, terms, numbers, parameters or the like

Below is a description of a method and system for selecting a background image from a video sequence with foreground objects. Area of the background in the frame that do not overlap the foreground objects during the video sequences can be captured by processing individual frames of the video sequence.

Figure 1 shows a specific non-limiting embodiment of the method of selecting the background image from the video sequence. In one embodiment, the method may include the localization of moving objects in the image using the mask change detection at the stage 110 and the model training background in other areas of the image on the stage 120. The localization of moving objects on the stage 110 boundaries of moving objects, which have a uniform color for at least two consecutive frames, labeled by creating one or more membranes that cover the area corresponding to the moving objects. The rest of the image is considered as the background and is used to train the model of the background on the stage 120. In one var is ante perform background can also be used for detecting and compensating camera movement on stage 130.

On figa and 2B shows two successive frames of the same video sequence. As an example, the stage 110 in figure 1 it is assumed that in a video sequence is represented by only one moving object 205 (for example, part of the walking man), which has a uniform color. In the frame 25 of the part of the walking person 205 may have changed position compared to their position in the frame 250. The difference between these two frames - frame 250 and the frame 255 of the image is an object or its parts, which is moved and which is shown in figs in the form of an image 209 change detection. For example, the left leg 261 man barely visible in the image 209 as the person takes a step the right leg 264, at the same time keeping the left leg 206 is essentially no motion on the floor. Thus, the left leg 262 of the person does not appear in the image 209 change detection. In contrast, the heel of the right leg 264 men climbed out of the frame 250 in the frame 255, and so it appeared in the picture 209 change detection.

The use of masks 219 change detection leads to labeling of only the edge of circuits 210, 211 and 212 of the moving zones 209 uniform color, not of the zones themselves fully, as shown in fig.2D. For example, the circuit 210 corresponds to the border around the torso, hands and outer parts of the legs of the object 205; path 210 which corresponds to the border around the inner parts of the legs of the object 205; and circuit 212 corresponds to the head and neck of a moving object 205. In the mask 219 change detection has far fewer pixels than the total number of pixels in the frame. The use of the algorithm for change detection for high-resolution images with the subsequent processing of the mask change detection for localization movement takes much less time than using the sophisticated technology of the scanner, similar to the optical flow.

All localize moving objects by applying a quick analysis of the associated components to the mask 219 change detection, which construct the shell 239 around the contour of each of the moving zone, as shown on file. For example, the sheath 220 create around circuit 210, the shell 221 - around circuit 211, and a shell 222 - around circuit 212.

Let Itis the image at time t, mt⊂Itis the set of pixels that match really moving objects, and Mt⊂It- set of pixels that belong to one of the shells. Localization means that Mtshould cover mt. In practice, if pixel p belongs to St=It-Mtit will match a static object with a high degree of certainty.

To locate a moving object is in the detection algorithm changes are applied to frames of the video (for example, to frames 250 and 255). In one embodiment can be used, for example, the detection algorithm changes described in "Introduction to three-dimensional computer vision" Emanuel Trucco and Alessandro Verri, Publishing house "Prentice Hall, 1998. As an alternative solution you can use other algorithms for change detection. In addition, the detection algorithm changes, you can choose based on the requirements of a particular application.

If, for any nthen the pixel is considered as moving, whereis the maximum change in consistently current average value, so that the model of the background pixels is considered trained. The threshold value isselected as the product σ(n)computed from a sequence of images of a static scene, where σ is the standard deviation of the normal distribution of color pixels in the case of one or more color channels. In one embodiment, the mask change detection marks the zone change noise and lighting in addition to the moving boundaries of zones of uniform colour. As mentioned above, to localize a moving object creates the shell of these zones, so that it contains the moving pixels and no fun is no, if possible, the static pixels.

A moving object is the accumulation zones of detection of changes in the current moment of time t. For simplicity you can assume that there is only one moving object. All the related components in the mask detection of changes and their contours are installed. In one embodiment, for getting rid of noise contours (e.g., circuit 231 on fig.2D) zone with the small area are filtered out. Then choose the path Withmaxwith the largest area (which corresponds to the object or its border), for example, the circuit 220 on fig.2D. Iterative design of shell N begin by joiningmaxwith other areas of the circuits (e.g. circuits 221 and 222). These other square contours represent other moving area of the moving object 205.

Figure 3 shows an embodiment of an iterative construction of the shell. At stage 120 for all polygons Withidesign their convex hull. The convex hull is the smallest convex polygon that contains one or more components of the moving areas. The convex hull of the contour Withidenote as Hiand the convex hull of the contour Cmaxas Nmax. At stage 320 find the index k, so that the Euclidean distance between the Hkand Hmaxis minimum:

k=arg min(dist(HiNmax )) and dk=min dist(HiNmax).

At stage 340 determines whether the convex hull within the minimum distances Dmaxthe convex hull Withmax(dkless than the threshold value of Dmax). If so, construct the convex hullaround the set of shells Hkand Hmaxat stage 350. If not, then repeat stage 340 for the next circuit stage 345. At stage 360 denoteand in step 370 to determine whether consideration of all the paths. Then everything is repeated from step 320 until alliwill not be counted. Otherwise, move on to stage 380. At stage 380 sets the moving area is equal to the most recent maximum contour (Mt=Nmax). The above stages can be combined for the case of multiple moving objects.

The quality of the above algorithm can be evaluated by using the two values. The first value is a conditional probability that the pixel is treated as moving under the condition that it really corresponds to the moving object:

P1=P(p∈Mt|R∈mt).

The second value is a conditional probability that the pixel is treated as moving under the condition that it is static: R2=P(p∈Mt|p∈It -mt), where Itis the image at time t, mtis the set of pixels in the Itthat correspond to moving objects, and Mtis the set of pixels in the Itthat have undergone a significant change in color in the last or last few frames.

P1should be as large as possible, while R2should be small. If P1is large enough, it can be executed training distorted background, while at small enough P2increased learning time. P1and R2should obviously increase with the increase of Dmax. This determines Dmaxas the minimum value that ensures P1above a certain level of confidence. The choice of Dmaxwill be described with reference to with regard to Fig.

As mentioned above, the mask change detection marks only the homogeneous boundary of the moving areas. In addition, it may not mark the areas that are moving quite slowly. Therefore, some slow-moving objects can constantly move in the background, and some moving objects can accidentally be considered as belonging to the background. One solution of the first problem consists in performing change detection multiple times with different reference frames, such as one to the Drome before the current frame, two frames before the current frame, etc. One solution of the second problem consists in performing training background taking into account the fact that some frames of the background can be distorted. In this regard, two features of the localization algorithm of motion are of interest: the probability of R(m)what a moving pixel is misclassified m times in a row, and the index m* is such that a probability of P(m*)is below the confidence level, in this case, m* can be used as a parameter for algorithm training background.

As shown in figure 1, when all the moving area in the current frame is localized on the stage 110, then the model training background with the specified static pixels of the current frame on the stage 120. The color of the pixel can be characterized at a given point in time three values {X(n)}, n=1...3, which in the case of static pixel acceptable can be modeled by normal distributions N(μ(n)that σ(n)) unknown medium μ(n)and standard deviations σ(n).

Training is performed in several stages to remove emissions generated incorrect prediction on the stage 110. With random changes of the background can be treated similarly. If the pixel in the foreground is a normal distribution with a small variance in those who tell long time, it is considered as changing the background and the model of the background immediately updated. For segmentation of the background in each image you can use background subtraction, as described, for example, in the article "Not a parametric model for background subtraction" Ahmed ElGamal, David Harwood, Larry Davis, Proc. ECCV, volume 2, pages 751-767, 2000. In an alternative embodiment, the present invention can use other methods of background subtraction.

During the learning process using the calculated values μ(n)using the pack current average:

where tiindicates the frames where the pixel was classified as static.

When a sequence converges, i.e. the difference betweenandis small:

the model background is considered trained in that pixel and. Therefore, each pixel can correspond to one of the four States shown in figure 4: state 410 of the unknown background (which corresponds to pixels that have never been in St), state 420 untrained background (when dialed statistics not performed inequality 2), state 430 trained background (the inequality 2) and state 440 foreground (the hen training background performed on the current frame detected foreground by subtraction of the background). Possible transitions are shown in figure 4. The transition A occurs when the pixel appears in Stfor the first time. The transition V occurs when the model pixel is considered as sufficiently trained. The transition S occurs when the foreground is static for a long period of time.

To simplify the pixel at a given time t can be characterized by using only one value of Xt. Equation (1) and inequality (2) contain unknown parameters α and βthat define the learning process. Suitable choice of these parameters provides a fast and at the same time statistically optimal training background. Assuming that X1=I+Δt, where I is a constant pixel color of the background, and Δ noise is Gaussian with zero mean color of a pixel at time t with a standard deviation of Δσfor δtt-I get the following equality δt=(1-α)δti-1+αΔtiwhere δtis the difference between the current average and a constant background color.

δthas a normal distribution with mean and standard deviation of σt.

where and is the current average constant

In order to have a reliable background education background on what should be a long enough time, to be sure that the training background is not performed using a moving object. In other words, if the pixel value changes significantly, then the training should last at least for m* frames. So should run the following inequality:

where δtowell σΔ and m* is the minimum number of consecutive frames, so that the probability of P(m*)below the level of reliability; in other words, we can assume that no pixels were correctly classified in all m* a sequential frames. In one embodiment, there may be no reason to do β less than the value set in the inequality (4), because it dramatically increases the time required for training the background.

At the same time, standard deviation δm*should be as small as possible. You can prove thatas function α∈[0, 1] has at least onewhere

Examples ζ(α) for different numbers of frames shown in figure 5.

Figure 5 shows an example of performing the relative scattering of the current average depending on the values α. In one embodiment, the solid line 510 corresponds to the fifth frame, the dotted line 520 corresponds as ten is the frame, and the dash-dotted line 530 corresponds to the twentieth frame.

The selection is too large or too small value and leads to large statistical uncertainties δand the current average μa=a*m*you can choose so that when a static pixel of the background current, average μtm*taken as a value of the background pixel has the minimum possible standard deviation. Given m*, the inequality 4 and equality 5 specify the optimal value β and α.

In one embodiment, the change of the background can be taken into account during model training background. Assume that the camera is not moving and the background has changed significantly, although this remains static. For example, one of the static objects moved to another position. The system marks the former and the current location of the object as the foreground. These pixels are usually not the foreground pixels, and are static in the background. This property allows you to track such changes and background to adapt the model of the background. The model is trained for each pixel in the background, and if he has a static behavior over a long period of time, then its status changes in status not trained background. After a given number of frames (e.g., three frames), it becomes a trained background.

As shown in figure 1, in one embodiment, the background can also what to use for detecting and compensating camera movement on stage 130. The methods above can be generalized for the case of a moving camera by enabling the rapid detection of global motion. When part of the image taking condition 430 trained background, according to figure 4, then subtract 450 background can be applied to each frame and the algorithm for global motion estimation affects found the mask of the background.

7 shows an embodiment of the detecting and compensating camera movement. In one embodiment, the characteristics of the selected frame with the ability to track background at stage 710, for example, angles 681-693, as shown in Fig.6. You can use optical flow to track several strong characteristics of the background to detect motion of the camera at the stage 720. In one embodiment, the technology choice of signs corresponds to the technology described in the article "Good features to track" Jianbo Shi and Carlo Tomasi, Proc. CVPR, pages 593-600, 1994. In one embodiment, the tracking technology of signs corresponds to the technology described in "Introduction to three-dimensional computer vision" Emanuel Trucco and Alessandro Verri, Publishing house "Prentice Hall, 1998. As an alternative solution you can use other features, the choice of features and ways to track.

After detection of the global motion in the background, indicating the movement of the Cam the market at stage 730, the model of the background is returned to the original state at stage 740 by transferring all pixels in a condition of unknown background (for example, state 410 figure 4). Tracking signs provides a good estimate of the global motion tracking points in a sustainable manner for a long time. If all pixels of the background is lost, it is possible to trace a certain percentage of the pixels of the detection algorithm changes. If you find a false end of the movement (may have a slight speed change is detected during the movement of the camera, for example, due to the uniform background), stages 110 and 120 localization movement and learning, according to figure 1, it filters out incorrect values of the pixels. When on stage 760 camera stops moving, then starts again at the model training background for each value of pixel (stage 120 figure 1).

Below are some results of experiments using the localization movement and training background. It should be noted that the experimental results are shown only for the purposes of a more visual description of the present invention and should not limit the invention. In one embodiment, the described scheme was implemented using a Library of image processing (Image Processing Library), available on the market by Intel and Libraries for computer vision with open source (OpenCV) Intel system capable of processing 320×240 images for 15 msec. Testing was performed on a large number of video sequences captured directly by the camera USB.

The threshold value Dmaxlocalization movement can in one embodiment be selected according Fig. On Fig shown as an example of the results of testing the algorithm on video sequences and compared these results with the foreground segmentation based on background subtraction. P-value1represents the percentage of pixels of the foreground, which was classified as the moving pixels. In alternative versions of the threshold value Dmaxyou can choose based on other empirical data or by using other methods, such as simulation, modeling and assumptions.

Figure 9 shows the percentage of background pixels that are segmented as foreground, obtained using the same methods. Above P1and R2can be adjusted using the parameter Dmax. For Dmax=15 calculates the number n(m) of foreground pixels that were incorrectly classified m times in a row. The results are shown in the following table:

M123456
N54232023812830

Taking m*=5, to get the above inequality (4) and equality (5) the following values: α˜0.25 and β˜0,71.

Figure 10 shows an embodiment of a computer system (e.g., client or server) in a digital processor system, representing, for example, a workstation, a server, a personal computer, a compact computer, laptop computer, personal digital assistant, a phone, a set top box, etc. in which it is possible to realize the characteristics of the present invention. Digital processor system 1000 can be used for applications such as video surveillance, videoconferencing, vision robots, etc.

Digital processor system 1000 includes one or more buses or other means to transfer data between components of a digital processor system 1000. Digital processor system 1000 also includes means for processing data, such as processor 1002 connected to the system bus for processing information. The processor 1002 may represent one or more General-purpose processors (e.g., processor, Motorola PowerPC, or p is ocessor Intel Pentium) or a dedicated processor such as a digital signal processor (DSP) (e.g., Texas Instrument DSP). The processor 1002 may be configured to execute commands to implement the above operations and stages. For example, the processor 1002 may be performed by processing algorithm for localization of a moving object in frames of the sequence.

In addition, the digital processor system 1000 includes a system memory 1004, which may include random access memory (RAM) or other device dynamic memory connected to the controller 1065 memory for storing data and commands to be performed by the processor 1002. The controller 1065 manages the transactions between the processor 1002 and memory devices such as memory 1004. The memory 1004 may also be used for storing temporary variables or other intermediate information during execution of commands by the processor 1002. The memory 1004 represents one or more memory devices, such as memory 1004 may also contain permanent memory (ROM) and/or other device static storage for storing static information and commands to the processor 1002.

Digital processor system 1000 can also include a controller 1070 input/output control operations between the processor 1002 and one or more devices 1075 I / o, keyboard and mouse. Counter is ller 1075 I / o may also control operations between the processor 1002 and peripheral devices, for example, device 1007 storage. Device 1007 storage represents one or more storage devices (e.g., a magnetic disk drive or optical disc drive)connected to the controller 1070 input-output, for storing data and commands. Device 1007 storage can be used to store commands to perform the above stages. The controller 1070 I / o can also be connected with the base system 1050 input / output system (BIOS) to load the digital processor system 1000.

Digital processor system also includes a video camera 1071 for recording and/or playback of video sequences. Luggage 1071 may be connected to the controller 1070 I / o using, for example, a universal serial bus (USB) 1073. As an alternative solution for camera connection 1071 controller 1070 I / o you can use other types of tires, such as fire wire bus. Controller 1070 I / o can be connected display device 1021, such as a cathode ray tube or liquid crystal display, for displaying video sequences for the user.

The device 1026 connection (such as a modem or an interface card and the network can also be connected to the controller 1070 IO. For example, the device 1026 may be - lo the existing Ethernet network, - LAN token-ring or other type of interface to provide a communication line network to which the digital processor system 1000 adapted to establish a connection. For example, the device 1026 connection can be used to receive data related to the sequences from the other camera and/or computer system or network.

It should be noted that shown in figure 10 architecture is an example only. In alternative versions of the present invention can use a different architecture for the digital processor system 1000. For example, the controller 1065 memory and controller 1070 I / o can be integrated into a single component and/or different components may be connected to each other in other configurations (for example, directly with each other or with other types of tires.

In the above description, specific examples introduced a new and quick way to extract the background from an image sequence with a moving foreground objects. The method uses the operation of the image processing and circuits and is able to reliably extract the background for a small number of frames. For example, the method can work with about 30 frames in a typical video sequence with a static background and one person on p is rednam plan. This is a significant advantage for videopremiere in real time, such as surveillance and vision robots, compared with systems known according to the prior art, which are based on expensive computational operations. Methods according to this invention can be applied to solve wide range of tasks that are associated with stationary background and interest objects in the foreground. Additionally, the versatility of the system allows for the selection of the detection algorithm changes for the needs of a particular application. Such methods can also be used in connection with the compression of the video data with the advantage of the knowledge of static zones in the sequence.

In the above description was presented to the invention with reference to specific examples of its implementation. However, it is obvious that various modifications and changes without departing from the more broadly understand the nature and scope of the invention presented in the annexed claims. In accordance with this description and the drawings should be considered as an illustration and not to limit the invention.

1. The way to select the background image from the video sequence, containing the following steps: localization of a moving object in a video sequence based on the change is of a moving object in multiple frames of the video sequence, when a moving object is the area of the frame changing color; and model training background of many shots outside the zones frames changing color.

2. The method according to claim 1, characterized in that the localization contains the localization of a moving object using a mask change detection.

3. The method according to claim 1, characterized in that the localization contains

the definition of the boundaries of a moving object, which has a uniform color and design of shell around a moving object using a border.

4. The method according to claim 3, characterized in that the determination of the boundary includes the following stages: definition of the maximum path from the multiple paths of a moving object, the maximum contour has the largest area among the multiple paths; identify other circuits of the moving object and combining maximum circuit with other circuits.

5. The method according to claim 4, characterized in that it further comprises the step exception of the smallest path from Association with a maximum contour.

6. The method according to claim 4, wherein the combining includes combining one of the other circuits with a maximum contour, if the distance between the maximum circuit and one of the other paths is less than a specified distance.

7. The method according to claim 6, characterized in that the frames contain multiple what about the pixels and the set distance based on the probability the pixel many pixels is considered as a moving, if it belongs to a moving object.

8. The method according to claim 7, wherein the specified distance is based on a probability that the pixel is treated as moving under the condition that it is static.

9. The method according to claim 3, characterized in that the frames contain many pixels, and a shell design so that it contains only the pixels with changing color in consecutive frames.

10. The method according to claim 3, characterized in that the construction of the shell includes the following steps: identify all connected elements in the border, in which each component is a path with a space; a filter circuit with the smallest area; selecting a path with the maximum area and merge circuit with a maximum size of with other circuits connected components.

11. The method according to claim 1, characterized in that the frames contain many pixels and training contains a characterization of the pixel color at the specified time value based on condition, however, each pixel corresponds to one state from multiple States.

12. The method according to claim 11, characterized in that the set of conditions includes the condition of untrained background.

13. The method according to claim 11, characterized in that the set of conditions includes the condition of the scientists of the background.

14. The method according to claim 11, characterized in that the set of conditions includes the condition of the foreground.

15. The method according to claim 11, characterized in that the set of conditions includes the condition of unknown background.

16. The method according to claim 11, wherein the training includes the following stages: model training background pixel in the foreground and the status of the pixel in the untrained state background when the pixel shows the static behavior within a certain period of time.

17. The method according to item 16, characterized in that it further comprises changing the state of a trained background after a specified number of frames.

18. The method according to claim 1, characterized in that the sequence record with your camcorder and the method further comprises the following steps: motion detection video cameras and motion compensation video camera.

19. The method according to p, wherein the motion detection includes the following steps: the choice of the sign frame and the tracking characteristics of the frame over multiple frames.

20. The method according to claim 19, wherein the compensation includes the reset model the background when the movement stops.

21. A machine-readable storage medium that is used to select the background image from the video sequence, having not recorded on the team, which when executed by a processor, cause the performance of the processor the following operations: localization of a moving object in a video sequence based on change of a moving object within the set of frames of a video sequence, while the moving object occupies the area of the frame changing color; and model training background of many shots outside of the frame changing color.

22. A machine-readable storage medium according to item 21, wherein the localization includes localization of a moving object using a mask change detection.

23. A machine-readable storage medium according to item 21, wherein the localization contains the following stages: defining the boundaries of a moving object, which has a uniform color; and construction of shell around a moving object using a border.

24. A machine-readable storage medium according to item 23, wherein determining the boundaries of the following phases: definition of the maximum path from the multiple paths of a moving object, the maximum contour has the largest area among the multiple paths; identify other circuits of the moving object and combining maximum circuit with other circuits.

25. A machine-readable storage medium according to paragraph 24, characterized in, is the processor additionally performs the following commands: define the smallest path from multiple paths and exception of the smallest path from Association with a maximum contour.

26. A machine-readable storage medium according to paragraph 24, wherein the combining includes combining one of the other circuits with a maximum contour, if the distance between the maximum circuit and one of the other paths is less than a specified distance.

27. A machine-readable storage medium according to item 23, wherein execution of the processor design shell includes execution by the processor following commands: identify all connected elements in the border, in which each component is a path with a space; a filter circuit with the smallest area; selecting a path with the maximum area and merge circuit with a maximum size of with other circuits connected components.

28. A machine-readable storage medium according to item 21, wherein the frames contain many pixels and the execution processor of the instruction includes executing processor characterization pixel color at the specified time value based on condition, however, each pixel corresponds to one state from multiple States.

29. A machine-readable storage medium according p, characterized in that the execution processor of the instruction includes executing by the processor the following commands: model training background pixel in the foreground and the change is Oceania pixel in the untrained state background if the pixel shows the static behavior within a certain period of time.

30. A machine-readable storage medium according to item 21, wherein the video sequence is recorded by the video camera and in which the processor additionally performs the following commands: motion detection video cameras and motion compensation video camera.

31. A machine-readable storage medium according to item 30, wherein execution by the processor of the motion detection includes performing by the processor following commands: select sign frame and track frame signs over multiple frames.

32. A machine-readable storage medium according to item 30, wherein execution by the processor compensation includes execution by processor reset model the background when the movement stops.

33. The device for selecting the background image from the video sequence that contains a tool for localization of a moving object in a video sequence based on change of a moving object within the set of frames of a video sequence, while the moving object occupies the area of the frame changing colors; and means for training the model background for many shots outside of the frame changing color.

34. The device according to p, characterized in that the means for local the organization includes means for determining the boundaries of a moving object, which has a uniform color; and means for constructing wrapper around a moving object using a border.

35. The device according to p, characterized in that the sequence is recorded by the video camera and in which the device further includes means for detecting movement of the camera and means for motion compensation of video.

36. The device for selecting the background image from the video sequence that contains the processor to execute one or more programs for localization of a moving object in a video sequence based on change of a moving object within the set of frames of a video sequence, while the moving object occupies the area of the frame changing color; and model training background of many shots outside of the frame changing color, and the storage device, coupled to the processor, the storage device has stored therein one or more programs for localization of a moving object and model training background.

37. The device according to p, wherein the processor is adapted to execute one or more programs for localization of a moving object using a mask change detection.

38. The device according to p, wherein the processor is adapted to execute one or more programs to determine the s border of a moving object, which has a uniform color and to construct a shell around a moving object using a border.

39. The device according to p, characterized in that it further comprises a display coupled to the processor, adapted to display multiple frames of the video sequence.

40. The device according to p, characterized in that it further comprises a camera connected to the processor, for recording a set of frames of a video sequence.

41. The device according to p, wherein the processor is adapted to execute one or more programs for motion detection camcorder to compensate for motion of the camera.



 

Same patents:

FIELD: television.

SUBSTANCE: support frame is assigned with sign, showing information about direction of support frame, and during determining of predicted vector of movement of encoded block averaging operation is performed with use of vectors of movement of neighboring blocks, during which, if one of aforementioned blocks has movement vectors, information about direction of support frames is received, to which these movement vectors are related, and one of movement vectors is selected with reference to received information about direction, than averaging operation is performed with use of selected movement vector to receive subject movement vector of encoded block.

EFFECT: higher precision, higher reliability.

3 cl, 1 dwg, 3 ex

The invention relates to a method and apparatus for identification and localization of areas with relative movement in the scene and to determine the speed and oriented direction of this relative movement in real time

The invention relates to the field of image processing and can be used in automated systems management traffic, for monitoring and documenting the landing maneuvers at airports, in robotics and in a more General approach can serve as a subsystem for systems with a higher level of interpretation, which are detected, segmented and can be observed moving objects, and automatically defined parameters

The invention relates to a video system technology and can be used when designing a digital coding device for video telephony, video conferencing, digital television broadcasting standard and high definition

The invention relates to a video system technology and can be used when designing a digital coding device for video telephony, video conferencing, digital television broadcasting standard and high definition

FIELD: technology for processing remote probing data for detection and recognition of objects on basis of their images.

SUBSTANCE: method includes source geometrically pixel-wise combined digitized images of one and the same scene concurrently in n spectrum ranges, source signs matrix is formed, each element of which represents n-dimensional values vector of signals of pixels of source images with similar coordinates, standard is selected in form of arbitrary element of source signs matrix, final numeric matrix is formed, to each current element of which value is assigned, equal to distance in vector signs space between vector, appropriate for standard and vector, appropriate for element of source matrix with same number of row and column to those of current element, final matrix is transformed to digital image, while as signs, brightness values of pixels are used, textural and gradient characteristics of source image pixels.

EFFECT: simplified operations concerning generation of synthesized image for visual interpretation, its adaptation to objects targeted by observer, detailed reflection of chosen objects on synthesized image and compact representation of information.

4 cl, 2 dwg

FIELD: optical object identification means.

SUBSTANCE: identification element of optical seal is lit by passing optical probing emission, its optical image is projected into place of multi-element photo-detector, voltages Un are recorded at outputs of elements of photo-detector and aforementioned signals are used to form an optical image of seal, which is recorded and then used to compare to control optical image of seal. During recording process aforementioned voltages are measured and also measured are exposition time spans appropriate for them in case of presence pot absence of identification element in optical track, and also voltages during replacement of identification element by black body and measured values are used for calculation by given method of a block of mathematical expressions, representing an optical image of seal.

EFFECT: optical image, free from influence of alternation of a parameters of optical track and photo-detector elements.

1 dwg

FIELD: communications.

SUBSTANCE: previously, at transmitting and receiving sides random quadratic matrix is identically generated with size m×m elements and two pairs of random key matrices with sizes N×m and m×N elements. From k frames of colored images with sound signal k of matrices of quantized counts of colored moving image are formed with size M×M×k elements and Z-digit sound vector. Received matrices are transformed to digital form on basis of presentation of each of them in form of result of multiplication of three matrices: random rectangular matrix with size N×m elements, random quadratic matrix with size m×m elements and random rectangular matrix with size m×N elements. Into digital communication channel elements of rectangular matrices of size N×m and m×N elements are transferred. Restoration of images is performed in reversed order.

EFFECT: higher speed of data transfer.

5 cl, 23 dwg

FIELD: text optical recognition.

SUBSTANCE: method includes following stages: separating data of bitmap image on levels with varying complication degree; separating these data on objects; determining membership of each objects at one of complication levels; setting hierarchical connections between objects of different complication levels; setting interconnection between objects of same complication level; analyzing properties of objects, including at least following steps: making a theory about properties of analyzed object, checking theory concerning properties of analyzed objects, correcting data concerning properties of connected objects of same or different complication levels.

EFFECT: higher quality, simplified processing, higher error sensitivity.

6 cl, 3 dwg

FIELD: coding elementary-wave data by means of null tree.

SUBSTANCE: proposed method includes generation of elementary-wave ratios pointing to image. In the process bits of each elementary-wave ratio are associated with different bit places so that each place is associated with one of bits of each elementary-wave ratio and associated bits are coded with respect to each place of bits to point to null tree roots. Each place of bits is also associated only with one of bits of each elementary-wave ratio. Computer system 100 for coding elementary-wave ratios by means of null tree has processor 112 and memory 118 saving program that enables processor 112 to generate elementary-wave ratios pointing to image. Processor 112 functions to code bits of each place to point to null tree roots associated with place of bits.

EFFECT: enhanced data compression speed.

18 cl, 7 dwg

The invention relates to computing, and in particular to control systems for the identification and conversion of formats of representations of text documents in the information network of the State automated system (GUS) “Election”

The invention relates to the field of optical character recognition of text from a bitmap

The invention relates to the field of telecommunications

The invention relates to forestry, remote methods of solving problems of forestry purposes

The invention relates to television and provides as technical result increased recognition accuracy is also displayed if any image on the screen of a television receiver

FIELD: coding elementary-wave data by means of null tree.

SUBSTANCE: proposed method includes generation of elementary-wave ratios pointing to image. In the process bits of each elementary-wave ratio are associated with different bit places so that each place is associated with one of bits of each elementary-wave ratio and associated bits are coded with respect to each place of bits to point to null tree roots. Each place of bits is also associated only with one of bits of each elementary-wave ratio. Computer system 100 for coding elementary-wave ratios by means of null tree has processor 112 and memory 118 saving program that enables processor 112 to generate elementary-wave ratios pointing to image. Processor 112 functions to code bits of each place to point to null tree roots associated with place of bits.

EFFECT: enhanced data compression speed.

18 cl, 7 dwg

FIELD: text optical recognition.

SUBSTANCE: method includes following stages: separating data of bitmap image on levels with varying complication degree; separating these data on objects; determining membership of each objects at one of complication levels; setting hierarchical connections between objects of different complication levels; setting interconnection between objects of same complication level; analyzing properties of objects, including at least following steps: making a theory about properties of analyzed object, checking theory concerning properties of analyzed objects, correcting data concerning properties of connected objects of same or different complication levels.

EFFECT: higher quality, simplified processing, higher error sensitivity.

6 cl, 3 dwg

FIELD: communications.

SUBSTANCE: previously, at transmitting and receiving sides random quadratic matrix is identically generated with size m×m elements and two pairs of random key matrices with sizes N×m and m×N elements. From k frames of colored images with sound signal k of matrices of quantized counts of colored moving image are formed with size M×M×k elements and Z-digit sound vector. Received matrices are transformed to digital form on basis of presentation of each of them in form of result of multiplication of three matrices: random rectangular matrix with size N×m elements, random quadratic matrix with size m×m elements and random rectangular matrix with size m×N elements. Into digital communication channel elements of rectangular matrices of size N×m and m×N elements are transferred. Restoration of images is performed in reversed order.

EFFECT: higher speed of data transfer.

5 cl, 23 dwg

FIELD: optical object identification means.

SUBSTANCE: identification element of optical seal is lit by passing optical probing emission, its optical image is projected into place of multi-element photo-detector, voltages Un are recorded at outputs of elements of photo-detector and aforementioned signals are used to form an optical image of seal, which is recorded and then used to compare to control optical image of seal. During recording process aforementioned voltages are measured and also measured are exposition time spans appropriate for them in case of presence pot absence of identification element in optical track, and also voltages during replacement of identification element by black body and measured values are used for calculation by given method of a block of mathematical expressions, representing an optical image of seal.

EFFECT: optical image, free from influence of alternation of a parameters of optical track and photo-detector elements.

1 dwg

FIELD: technology for processing remote probing data for detection and recognition of objects on basis of their images.

SUBSTANCE: method includes source geometrically pixel-wise combined digitized images of one and the same scene concurrently in n spectrum ranges, source signs matrix is formed, each element of which represents n-dimensional values vector of signals of pixels of source images with similar coordinates, standard is selected in form of arbitrary element of source signs matrix, final numeric matrix is formed, to each current element of which value is assigned, equal to distance in vector signs space between vector, appropriate for standard and vector, appropriate for element of source matrix with same number of row and column to those of current element, final matrix is transformed to digital image, while as signs, brightness values of pixels are used, textural and gradient characteristics of source image pixels.

EFFECT: simplified operations concerning generation of synthesized image for visual interpretation, its adaptation to objects targeted by observer, detailed reflection of chosen objects on synthesized image and compact representation of information.

4 cl, 2 dwg

Up!