Method for processing signals for selecting moving objects in a series of television images |
|
IPC classes for russian patent Method for processing signals for selecting moving objects in a series of television images (RU 2311684):
                                                                                                                                                                                                     
 Method (variants) and device for identification of digital video data from several sources, television surveillance system / 2310289
In accordance to the invention, a set of digital video data from first source, representing first image, is identified as standard video data of first source. Then a second set of video data is read, which represents current image. Difference ratio is computed using standard digital video data and current set of digital video data. If difference ratio exceeds a threshold, a query is shown to system user on the display to receive a response about whether current digital video data belong to identified source, or originate from new source. If response points at new source, then current set of digital video data is dispatched for storage into second memory cell, connected to second source. Then the current set of digital video data is identified as standard digital video data for second source.
 Television system for monitoring movement of hot rolled metal / 2299524
In accordance to invention, components of combined image: with increased scale and normal (without scaling), - are formed in charge form on targets of first and second television sensors at varying times of exposition, optimal for each one of scene fragments being transmitted. Due to that, in output image of television system bright and light sections (hot rolled metal, measuring ruler) are transferred same as in prototype, without limitations of white, while dark and/or low light sections are transferred with high signal/noise ratio, and dynamic range of brightness levels is expanded.
 Device and method for automated control of situation in auditoriums / 2296434
In accordance to invention at least one video camera is mounted in auditorium for producing image of auditorium, and at least one computer with memory, interconnected via a local area network, while in memory of computer a database is generated, storing data, reflecting filling of auditorium, in accordance to sold tickets, and computer can process video signal received from video camera for producing data about filling of auditorium with possible storage of these data in memory for following analysis, and also with possible comparison of produced data to data stored in database and with possible generation of signal of disruption of set mode in case if mismatch of data being compared exceeds the predetermined threshold value.
 System to prevent accidents on railway tracks / 2295470
Invention relates to safety devices to be installed on dangerous section of railways. Observation and warning equipment with video processor subsystem for processing and recording images, alarm signaling unit and transmitting part of wireless communication subsystem is installed in area of potential dangerous section of track, for instance, crossing area. Train is furnished with train movement inter locking equipment, sound annunciator, video monitor and receiving part of wireless communication subsystem. Said video processor subsystem includes two video cameras covering potentially dangerous section of track. It includes also video image processor, image recorder, timer, video camera control unit, illumination pickup, lighting unit and resolver. Transmitting and receiving parts of wireless communication subsystem are made in form of video image transmitter and receiver with voice accompaniment. Display of video monitor is in field of vision of driver. Train movement inter locking equipment is operated by driver. System provides driver with on-line information on situation at nearest dangerous zone of crossing. Driver takes decision on emergency braking of train or continuation of movement basing of information available.
 Multichannel video-audio surveillance method and integrated high-frequency system for realization of said method / 2250504
Method includes surveillance of state of object, by surveillance blocks, each of which includes camera and microphone, low-frequency signals received from each block are converted to high-frequency television modulated signals, which are inputted into unified cable main, formed by coaxial television cable, along which received independent signals are sent to input of control panel, in form of television receiver or computer, provided with extension board, allowing to receive and display images of several surveillance objects concurrently, while power is fed along coaxial chamber of television cable main.
 Method for multichannel video-audio surveillance and system for realization of said method (variants) / 2250503
Method includes video surveillance of controlled object state, while into television cable main of object high-frequency television modulated signal is sent, while to receive signal concerning state of S objects, each of which includes group of N video surveillance blocks, including camera and microphone, video-audio signals from each group of N video surveillance blocks are combined along low frequency, received complex video signal is converted from each group of N video surveillance blocks into high-frequency television modulated signal and it is synchronized with unified cable main - coaxial television cable, in arbitrary groups combination, via which received independent S signals are sent to input of visualization and/or recording systems.
 Method of detecting objects / 2250478
Method comprises subtracting reference and current images, breaking the image series to be processed into fragments, and converting the characteristic features of the images into signals. The signals from one of the images are recorded as reference ones and are compared, e.g., by subtracting, with corresponding current signals, and, after the threshold processing, the difference signals obtained are converted into the binary signals for control of spatial filtration . As a result, the fragments of the current image, for which the control signals exceed the threshold, are transmitted, whereas the fragments, for which the signals are equal or less than the threshold value, are suppressed.
 Opto-electronic system review and maintenance / 2237979
The invention relates to the field of devices which are placed on a movable base opto-electronic devices that convert electromagnetic radiation into an electrical signal that carries information about the image, and videosmotorola device for monitoring process
 Closed-circuit television system "day-night" / 2234818
The invention relates to a television technique, namely surveillance, detection, differentiation and identification of dynamic objects with-the-clock work
 The way the visual spectral analysis of the television image of the far infrared range and device implementing this method / 2233559
The invention relates to television systems, particularly to television systems with cameras far infrared range
 Space-time prediction for bi-directional predictable (b) images and method for prediction of movement vector to compensate movement of multiple images by means of a standard / 2310231
The method is claimed for usage during encoding of video data in video encoder, containing realization of solution for predicting space/time movement vector for at least one direct mode macro-block in B-image, and signaling of information of space/time movement vector prediction solution for at least one direct mode macro-block in the header, which includes header information for a set of macro-blocks in B-image, where signaling of aforementioned information of space/time movement vector prediction solution in the header transfers a space/time movement vector prediction solution into video decoder for at least one direct mode macro-block in B-image.
 Method, device and software product for three-dimensional modeling of geological volume by selecting three-dimensional parameters of geological area / 2306607
In accordance to the invention, a set of multi-sided macro-cells (M) is determined, adapted to geometry of geological layers of the volume subject to modeling and geological space parameters are determined for determining match between geological area under examination and parametric area, by connecting a display point located in the parametric area to a point belonging to the geological area. Then, virtual division of multi-sided macro-cells (M) onto six-sided micro-cells (m), geometry of which results from division of geometry of each macro-cell (M), is determined.
 Device for tracking movement of mobile robot and method for same / 2305914
Suggested device for tracking movement of mobile robot includes: video camera for filming an individual object; unit for tracking movement and creation of image for setting support one in an image for current moment by means of filming of individual object by video camera and creation of image in current moment, for which support zone is set; unit for selecting image of difference of pixels of image support zone limit based on difference between pixels present only at limit of support zone of aforementioned images; and micro-computer for tracking movement of separate object on basis of selected image of difference.
 Method for determining movement vector being predicted / 2298886
In accordance to method, at least one movement vector is produced for at least one block, different from current block, while aforementioned at least one block is related to one, at least, supporting frame in a row of supporting frame, movement vector is predicted for current block on basis of received one, at least, movement vector, while prediction operation includes also operation of comparison of value of order number of B-frame to value of order number of one, at least, supporting frame, while movement vector for current block and aforementioned one, at least, movement vector are vectors of forward movement.
 Method for stereological examination of objects structural organization / 2291488
Method for examination of objects spatial organization is based on the following stages. Objects are subjected to stereological probe. Sizes, orientation, and/or location of received profiles of stereological probing are measured. Data arrays for examined objects are formed using measurement results. Array data is converted to statistical distribution of location coordinates for stereological probing profiles of objects. Obtained distributions are approximated by model distributions calculated for defined object parameters and stereological probing parameters.
 Method for graphic display of objects, moving in scenic space / 2284576
Method includes inserting enumeration system for each object and performing projection of enumerated objects onto plane, while projection is displayed in form of graph with trajectories of movement of enumerated objects in each staging.
 Method for evening out a grid of nodes in accordance to characteristic features in digital image / 2282242
Result is achieved by means of optimization and correction of grid nodes to provide extreme of compound function of spatial coordinates of grid nodes. Grid nodes are interpreted as atoms. Each node of grid provides a potential function for potential field of atom. Image represents a potential field. Compound function is a weighted total of potential fields of atoms and image, estimated in nodes of grid.
 Method for recognition of gestures in a series of stereo frames / 2280894
Method includes producing a series of stereo-images of object, on basis of which map of differences in depths is formed. System is automatically initialized on basis of probability model of upper portion of body of object. Upper portion of body of object is modeled as three planes, representing body and arms of object and three gauss components, representing head and wrists of object. Tracking of movements of upper part of body is performed with utilization of probability model of upper part of body and extraction of three-dimensional signs of performed gestures.
 Method of contact-free measurement of objects having defocused borders onto image / 2280838
Method of contact-free measurement of objects which have defocused borders onto image is based upon registration of object's image in memorizing unit, on presetting of rectangular areas of images fro subsequent detection of object's shape, on performing of differential-integral transforms, upon finding of coordinates of shape and calculating of sizes. In addition due to calibration of measurement system, the width of shape's lines is found depending on shift of object. While measuring, width of shapes of lines is found, and using the found dependence, the width of lines is transformed into distance from borders of object to defocusing plane, which distance is taken into account when calculating sizes of objects. Set of blanks with different thickness can be measured without re-focusing of system. Three-dimensional object sizes can be measured when faces of objects are disposed at some distance from focusing plane and are disposed not in parallel to it.
 Method and device for background segmentation on basis of movement localization / 2276407
Method contains localization of moving objects in each frame and learning of background model with utilization of image remainder.
 Objects recognition and tracking system / 2251739
System has matrix sensors, each of which is meant for performing functions of first type sensor, providing for possible detection of object presence in working zone of sensor and determining position thereof, and second type sensor, providing for possible use of this object position, determined by first type sensor, for identification or recognition of object, and possible focusing or operation with greater resolution then first type sensor.
|
FIELD: digital image processing technology, in particular, processing of signals for selecting moving objects in a series of television images. SUBSTANCE: in accordance to the invention, image turning angle of previous frame is determined relatively to standard image, increase of precision of calculation of shift parameters up to shares of pixel, change of standard image depending on computed values of shift and turn, shift of background image for integer number of pixels, turn of image around current frame around image center and following shift of turned image for fractional number of pixels, computation of value of threshold value with consideration of turbulence of atmosphere, vibration of image sensor and error when determining parameters of shift and turn, inter-frame filtration of threshold processing results. EFFECT: increased precision of object selection due to resistance to spatial distortions. 4 cl
The invention relates to digital image processing and can be used in security systems, vision systems, the system of space monitoring of the Earth and other A method of processing signals for selecting objects observed in the sequence of television images [U.S. Pat. Of the Russian Federation No. 2193825, IPC 04N 7/18, 2002], which is that of the image signals after analog-to-digital conversion is allocated 2N landmarks (N=3, 4, 5,...), used to estimate the parameters of the shift and rotate the current image to the previous one, and then select objects by thresholding the differential image obtained by subtracting from the signal of the current image of the image signal of the reference background, and the value of the threshold depends on the local scattering parameter differential image. The disadvantages of this method is the low accuracy of the estimates of the parameters of the shift and rotate the current image and the lack of compensation of the influence of the error of the estimate of the parameters of shear and rotation on the accuracy of selection of objects. Closest to the claimed method is chosen as the prototype of a way of automatically extracting signals of moving objects in image sequence in the presence of the geometrical is such distorted images [Alpatov B.A., Babayan PV Selection of moving objects in terms of the geometric distortion of the image // Digital signal processing. - 2004. No. 4. - p.9-14], consisting of analog-to-digital conversion of the image signal of each frame, remembering the first frame of the processed sequence as the reference image, remembering the first frame of the processed sequence as a background image, the determination of the parameters of the shift of the image of the current frame relative to the reference image with the precision of an element of the spatial resolution (pixel), the shift of the background image according to the calculated values of the shift parameters, pre-filtering of the background image within the first NCRframes, threshold processing of the absolute value of the difference signals of the image of the current frame and the background image for all subsequent frames, starting with the (NCR+1)-th (threshold value is determined by the variance of the additive noise on the image of the current frame), the recursive refinement of estimates of the brightness of the pixels of the background and the variance of the additive noise based on the results of the threshold processing. The disadvantage of this method is the impossibility to work in the shifts of the current frame relative to the first, is comparable in magnitude with the size of the image is, and when the movement of the image sensor in space gives rise to a more complex distortion of the image signal of the current frame than a simple shift relative to the reference background image (for example, a set of shift and rotation). An object of the invention is to improve the accuracy of feature extraction at the expense of resistance to spatial distortions, which have the form of shift and rotate the image of the current frame with respect to the previously observed frames. Technical result achieved in the implementation of the claimed invention is an analog-to-digital conversion of the image signal of each frame, remembering the first frame of the processed sequence as the reference image, remembering the first frame of the processed sequence as a background image, the determination of the parameters of the shift of the image of the current frame relative to the reference image, the offset of the background image in accordance with the obtained estimates of the parameters of the shift, pre-filtering of the background image within the first NCRframes, threshold processing of the absolute value of the difference signals of the image of the current frame and the background image for all subsequent frames, starting with the (NCR+1)-th recursive refinement of the estimates I have the bones of the dots of the background image and the variance of the additive noise based on the results of the threshold processing, when the rotation angle of the image of the current frame relative to the reference image determined to define the parameters of the shift of the image of the current frame relative to the reference image, calculating parameters of shift of the image of the current frame relative to the reference image is performed with sub-pixel accuracy, exceeding the calculated values of shear and rotation angle of the image of the current frame relative to the reference image maximum allowable value includes changing the reference image, after checking the conditions of a change of the reference image to calculate the parameters of the shift of the image of the current frame relative to the background image and rotation image of the current frame relative to the image of the first frame on detected values of the shift and rotation of the current frame relative to the reference image, shift the background image exercise on an integer number of pixels, after shifting the background image of the integer pixels perform the rotation, and the sub-pixel shift of the image of the current frame, when the threshold processing of the absolute value of the difference signals of the image of the current frame and the background image is the threshold value is calculated with consideration of atmospheric turbulence, vibration of the image sensor and netochno and define the parameters of the shift and rotation, after thresholding the absolute value of the difference signals of the image of the current frame and the background image additionally perform interframe filtering results threshold processing. The rotation angle of the image of the current frame relative to the reference image ϕTEyou can define rule
where F-1operator inverse discrete Fourier transform;
SSe(ωρthat ωϕ) - spectrum signal
(ρ, ϕ) - polar coordinates; ew(i, j) is the result of multiplying the centered signal from a reference image Δe(i, j) and the window function w(i, j)
ωρthat ωϕ- spatial frequency for the polar coordinate system. The shift parameters of the background image (αRthat βR) can be calculated as αR=round(αTFand βR=round(βTF), where round(...) is the function of rounding to the nearest integer value, αTF- shift the image of the current frame relative to the background image on the vertical axis, βTF- shift the image of the current frame relative to the background image on the horizontal axis, and the shift and rotate the image of the current frame can be defined as - ϕT1and (Δα, Δβ), where ϕT1- angle image of the current frame relative to the image of the first frame Miked the new filter results threshold processing can be performed according to the rule
where qn(i, j) is the result of interframe filtering in the form of a binary image
K - the number of analyzed frames preceding the current;
W is the sliding window size, placed sequentially in each point of the current frame (N=3, 5, 7,...); D - the minimum required number of frames for a decision on the membership of the considered point of the current frame object; n=2, 3, 4,... is the frame number. Thus, the differences of the proposed method from the prototype are as follows: 1) the presence of missing the detection phase angle image of the current frame relative to the reference image ϕTE; 2) improve the accuracy of calculation of the shift parameters αTEand βTEsub-pixel; 3) the stage of change of the reference image, based on the calculated values of shear αTEthat βTEand turn ϕTE; 4) missing the step of calculating the parameters of the shift αTFthat βTFand turn ϕT1image of the current frame relative to the background image on detected values αTEthat βTEand ϕTE; <> 5) shift the background image on an integer number of pixels (αRthat βR);6) the presence of a new phase of the rotation image of the current frame ln(i, j) on the corner - ϕT1around the center of the image and the subsequent shift the rotated image on a fractional number of pixels (Δα, Δβ); 7) use the new rules for determining the threshold value T, taking into account, in addition to the variance of the additive noise, air turbulence, vibration of the image sensor and the inaccuracy of determining the parameters of shift and rotate; 8) the stage interframe filter results threshold processing (binary image). The method of signal processing for selecting moving objects in a sequence of television images is as follows: 1) analog-to-digital conversion of the image signal of each frame of the observable sequence. 2) the memory image of the first frame of the processed sequence as the reference image; 3) remembering the image of the first frame of the processed sequence as a background image. 4) determination of the rotation angle of the image of the current frame relative to the reference image ϕTE; 5) determination of the parameters of the shift of the image is the current frame relative to the reference image α TEand βTEwith sub-pixel accuracy, where αTE- shift the image of the current frame relative to the reference image on the vertical axis, βTE- shift the image of the current frame relative to the reference image along the horizontal axis; 6) change the reference image when performing at least one of conditions (|αTE|>Tαβ), (|βTE|>Tαβ) and (|ϕTE|>Tϕ), where Tαβand Tϕrespectively the maximum values of shear and rotation; 7) calculation of shift parameters αTFthat βTFand turn ϕT1on detected values αTEthat βTEand ϕTE,where αTF- shift the image of the current frame relative to the background image on the vertical axis, βTF- shift the image of the current frame relative to the background image on the horizontal axis, ϕT1- angle image of the current frame relative to the image of the first frame; 8) shift the background image in (αRthat βR) pixels, where αRand βRis the result of rounding αTFand βTFto an integer value; 9) rotate the image of the current frame ln(i, j) on the corner - ϕT1around the center of the image and the subsequent shift the rotated image on a fractional number of pixels (Δ α, Δβ), where Δα=αTF-αRand Δβ=βTF-βR; 10) pre-filtering the converted background image within the first NCRpersonnel; 11) thresholding the absolute value of the difference signals converted image of the current frame and the transformed background image for all subsequent frames, starting with the (NCR+1)-St; 12) interframe filtering results of thresholding; 13) recursive refinement of estimates of the brightness of the pixels of the background and the variance of the additive noise based on the results of the threshold processing. The result of the analog-to-digital conversion of the image of each frame has the form of a matrix of numbers ln(i, j) In accordance with the developed method, the reference image is used to compensate for the observed spatial distortion of the image and the background image for extracting signals of moving objects. The processing of each frame of the observed image sequence, starting from the second, beginning what is to calculate the angle ϕ TEimage of the current frame ln(i, j) relative to the reference image e(i, j). The calculation is carried out according to the following rule, using the invariance property of the absolute value of the spectrum image from the image shift and the relationship between the rotation angle of the image in the Cartesian coordinate system and the amount of shift of the image in the polar coordinate system:
where F-1operator inverse discrete Fourier transform;
SSe(ωρthatωϕ) - spectrum signal
(ρ, ϕ) - polar coordinates; ew(i, j) is the result of multiplying the centered signal from a reference image Δe(i, j) and the window function w(i, j)
ωρthat ωϕ- spatial frequency for the polar coordinate system. The calculation of the spectra of the signals is performed using the algorithm of the fast Fourier transform. Multiplying the image signal by the window function w(i, j), falling to zero from the middle to the edges of the image, is performed to compensate for the Gibbs phenomenon. The processing of the first frame is ϕTEis assumed to be zero. Determination of the angle of rotation in accordance with the above rule allows to significantly improve the accuracy of estimation of the parameters of shear and rotation in comparison with the method proposed in [1]. Experimental studies have shown that when the signal-to-noise ratio (the ratio of contrast of the object to the standard deviation of the additive noise) is about 8, the maximum angle at which orota 10° and the total area of moving objects, not exceeding 10% of the total area of the image, the error of estimating the rotation angle does not exceed 10-3degrees. Definition of shift parameters (αTEthat βTE) image of the current frame relative to the reference carry out rule
where
Se(ωithat ωj) spectrum centered signal from a reference image Δe(i, j); ωithat ωj- spatial frequency for the Cartesian coordinate system. Sub-pixel precision of shift αTEand βTEis achieved by using a parabolic interpolation of the values array In order to maintain acceptable accuracy determination parameters αTEthat βTEand ϕTEthere is a change of the reference image each time, as calculated in the current frame, the value of any of the parameters exceed the corresponding maximum allowable value. Experimental studies have shown that when the signal-to-noise ratio of about 8, the maximum shift, not to exceed 15% of the linear dimensions of the image and the total area of moving objects, not exceeding 10% of the total area of the image, the error of estimating the shift does not exceed 0.5 element resolution (pixel). On the basis of the calculated parameter values αTEthat βTEand ϕTEdetermine the values of the parameters αT1that βT1and ϕT1linking the image of the current frame ln(i, j) and the image of the first frame in accordance with the expression αT1=αE1·cos(ϕTE)-βE1·sin(ϕTE)+αTE, βT1=βE1·cos(ϕTE)+αE1·sin(ϕTE)+βTE, ϕT1=ϕTE+ϕE1 where αT1- shift the image of the current frame ln(i, j) relative to the image of Pervov the frame on a vertical axis; βT1- shift the image of the current frame ln(i, j) relative to the image of the first frame on the horizontal axis; ϕT1- angle image of the current frame relative to the image of the first frame; αE1- shift the reference image e(i, j) relative to the image of the first frame on a vertical axis; βE1- shift the reference image e(i, j) relative to the image of the first frame on the horizontal axis; ϕE1- the angle of rotation of the reference image e(i, j) relative to the image of the first frame. Every time you change the reference image settings αE1that βE1and ϕE1update: αE1=αT1that βE1=βT1and ϕE1=ϕT1. Before the first shift of the reference image αE1=0, βE1=0 and ϕE1=0. Geometric transformation of the image of the current (nth) frame ln(i, j) and a background image ƒn-1(i, j)obtained by processing the (n-1) previous frames, exercise to compensate for the observed spatial distortion of the image of the current frame. In the points of both images with the same coordinates (indices) correspond to one and the same point of the observed scene. To perform geome the historical transformations of images l n(i, j) and ƒn-1(i, j) pre-compute the shift parameters αTFand βTFlinking the image of the current frame ln(i, j) and the background image ƒn-1(i, j), according to the rules αTF=αT1-αF1and βTF=βT1-βF1, where αTF- shift the image of the current frame ln(i, j) on the background image ƒn-1(i, j) along the vertical axis; βTF- shift the image of the current frame ln(i, j) on the background image ƒn-1(i, j) along the horizontal axis;
round(...is a function of rounding to the nearest integer value. After calculating the parameters αTFand βTFcalculate values αRand βRfor the current frame in accordance with the expression αR=round(αTFand βR=round(βTF). Compensation of spatial distortion is achieved by performing the following operations on signals of image: a) shift the background image ƒn-1(i, j) to (αRthat βR) pixels; b) rotate the image of the current frame ln(i, j) on the corner C) shift the image As a result of rotation and a sub-pixel shift of the image point with some coordinates (i, j) becomes a point of the resulting image, the coordinates of which are non-integer number, i.e. not coincide with the nodes of a discrete grid It is known that spatial interpolation leads to a weakening of the high-frequency component of the image signal (blurred contours, reduce the intelligibility of small parts of the image). It was established experimentally that the weakening of the high-frequency component of the background image, resulting from spatial interpolation leads to noticeable when iginio accuracy of selection signals of moving objects. Therefore, in accordance with the procedure for the compensation of the spatial distortions of the background image not subjected to operations that require spatial interpolation image rotation and sub-pixel shift. Shift the background image on an integer number of pixels (αRthat βRallows for the subsequent stages of processing the observed image sequence to take into account the background image gradual withdrawal from the field of view of the image sensor of some previously observed and the emergence of new plots of the observed scene, caused by the movement of the image sensor. The result of shifting the background image in (αRthat βR) pixels is
(the brightness of the pixels that correspond to previously not observed points in the image of the current frame is equal to the brightness of points in the observed image). At the stage of preliminary filtration assessment brightness ƒn(i, j) and the second initial moment μn(i, j) of pixels of the background for NCRframes in accordance with the expression
and
where
One is temporarily appreciate ƒ n(i, j) and μn(i, j) for each pixel of the observed image counts the number of frames kn(i, j)for which this point was present in the field of view of the image sensor. When processing the first frame of each point of the background image is mapped to a value of kn(i, j)equal to 1. When processing subsequent frames, the value of kn(i, j) is incremented for all observed points of the background image based on the shift of the background image in (αRthat βRpixels:
If for any point (i0I , j0) count the number of frames kn(i0I , j0) reached values of NCRit is considered that the obtained reasonably accurate estimates of the brightness and the second initial moment of the background image at the current point. Estimation of the variance of the additive noise dn(i, j)used in the threshold processing for extracting signals of moving objects, evaluated once only for those pixels of the background, which accumulated a fairly accurate estimate of the brightness and the second entry point, according to the rule dn(i, j)=μn(i, j)-[ƒn(i, j)]2when kn(i, j)=NCR. When processing of all subsequent frames calculated assessment of dn(i, j) at the points for which the k n(i, j)>NCRsubjected to the recursive refinement and shift in accordance with the calculated parameters (αRthat βR). On the threshold processing is the selection of moving objects by comparing the brightness of the image points of the current frame with the brightness of corresponding pixels of the background image. If the absolute value of the difference between two images in the current point (i, j) has exceeded the threshold value T, then accepted the hypothesis set of the considered point to the image of a moving object. When determining the threshold value T takes into account the variance of the additive noise, air turbulence, vibration of the image sensor and the inaccuracy of determining the parameters of shear and rotation. The result of threshold processing has the form of a binary image bn(i, j), computed according to the rule
where λ - width (1-p_)·100% confidence interval for the normalized Gaussian random variable; p_ - allowable probability of false classification point of the background to the object;
Experimental studies have shown that when the signal to noise ratio from 2 to 6 using the developed method allows to increase the frequency of correct selection of 5-10% or to reduce the frequency of false selection about 1.5-2 times compared with the prototype. Interframe filtering results threshold processing can significantly reduce the number of points, mistakenly attributed to the object or to the background. The result of the interframe filtering, as well as the result of thresholding is a binary image qn(i, j), computed according to the rule
where - Quantity of the Academy of Sciences of literami frames, preceding the current;
W is the sliding window size, placed sequentially in each point of the current frame (N=3, 5, 7,...); D - the minimum required number of frames for a decision on the membership of the considered point of the current frame object. The use of interframe filtering results of threshold processing by the signal to noise ratio from 2 to 5 allows you to increase the frequency of correct selection, on average, 30-50% while maintaining the same frequency of false selection. The recursive procedure to Refine estimates of the brightness of the pixels of the background ƒn(i, j) and the variance of the additive noise dn(i, j)obtained by pre-filtering the background image, allows to take into account changes in these parameters from frame to frame. Refinement of estimates ƒn(i, j) and dn(i, j) is carried out in accordance with the expression
The point of the background image, for which qn(i, j)=1, are unobservable in the current frame (i.e. closed moving objects), so the refinement of the estimates at these points is not performed. In pixels of the background image where the value of kn(i, j) has not reached the value of NPR ongoing assessment brightness ƒn(i, j) and the second initial moment μn(i, j) in accordance with expressions similar to the above (for stage pre-filtration). Count the number of frames kn(i, j), during which there was one or another point of the background image, continues to be updated and after the stage of preliminary savings estimates based on the results of image processing of the current frame according to the rules
If the image processing of the current frame counter value of the number of frames in the current point reached NCRthen, starting from the next frame, this point will be performed thresholding the absolute value of the difference image The proposed method of signal processing for selecting moving objects in a sequence of television images may be implemented on a personal computer (PC) utility engaged in the processing sequence of images coming from digital video or analog video camera through charge capture image (framegrabber). PR is using a PC Pentium IV 3 GHz (RAM - 512 MB, FSB - 4×187 MHz, operating system - Windows 2000 Professional or Windows XP) and the framegrabber Matrox Meteor II rate is 15 frames per second. There is also a variant of implementation of the proposed method is based on the PC General purpose, strict requirements to the frequency of the processing sequence of television images and different from the above-described variants of the fact that one of the PC is only used for solving the problem of compensation of spatial distortion, the other for selecting objects. While similar to the configuration of both the PC (Pentium IV 3 GHz, RAM 512 MB, FSB - 4×187 MHz, operating system - Windows 2000 Professional or Windows XP), the treatment frequency is 25 frames per second. When using the PC General purpose impossible (for example, when developing on-Board video systems for aircraft) or need a higher frequency processing, the proposed method of signal processing can be implemented on a programmable logic integrated circuits (FPGA), or the joint use of FPGA and specialized digital processing of signals, carries out the overall control of the computation process. Thus, the use of the proposed method in security systems technical SREN what I the system of space monitoring of the Earth and others will greatly improve the stability of such systems to the spatial distortion of the image caused by the movement of the image sensor, vibration and turbulence of the atmosphere compared to previously used methods. 1. The method of signal processing for selecting moving objects in a sequence of television images, consisting of an analog-to-digital conversion of the image signal of each frame, remembering the first frame of the processed sequence as the reference image, remembering the first frame of the processed sequence as a background image, the determination of the parameters of the shift of the image of the current frame relative to the reference image, the offset of the background image in accordance with the obtained estimates of the parameters of the shift, pre-filtering of the background image within the first NCRframes, threshold processing of the absolute value of the difference signals of the image of the current frame and the background image for all subsequent frames, starting with the (NCR+1)-th recursive refinement of estimates of the brightness of the pixels of the background and the variance of the additive noise based on the results of the threshold processing, characterized in that the angle of the current to the DRA relative to the reference image determined to define the parameters of the shift of the image of the current frame relative to the reference image, calculation of parameters of shift of the image of the current frame relative to the reference image is performed with sub-pixel accuracy, exceeding the calculated values of shear and rotation angle of the image of the current frame relative to the reference image maximum allowable carry out the change of the reference image, after checking the conditions of a change of the reference image to calculate the parameters of the shift of the image of the current frame relative to the background image and rotation image of the current frame relative to the image of the first frame on detected values of the shift and rotation of the current frame relative to the reference image, shift the background image is carried out on an integer number of pixels, after shifting the background image of the integer pixels perform the rotation, and the sub-pixel shift of the image of the current frame, when the threshold processing of the absolute value of the difference signals of the image of the current frame and the background image is the threshold value is calculated with consideration of atmospheric turbulence, vibration of the image sensor and the inaccuracy of determining the parameters of shear and rotation, after thresholding the absolute value of the difference signals of the image of the current frame and the background image additionally perform interframe f is litraly result of threshold processing. 2. The method according to claim 1, characterized in that the rotation angle of the image of the current frame relative to the reference image ϕTEdetermined by the rule:
F-1operator inverse discrete Fourier transform;
SSe(ωρthatωϕ) - spectrum signal
(ρ, ϕ) - polar coordinates; e w(i, j) is the result of multiplying the centered signal from a reference image Δe(i, j) and the window function w(i, j)
ωρthat ωϕ- spatial frequency for the polar coordinate system. 3. The method according to claim 1, characterized in that the shift parameters of the background image (αRthat βR) is calculated as αR=round(αTFand βR=round(βTF), where round(...) is the function of rounding to the nearest integer value, αTF- shift the image of the current frame relative to the background image on the vertical axis, βTF- shift the image of the current frame relative to the background image on the horizontal axis, and the shift and rotate the image of the current frame is defined as - ϕT1and (Δα, Δβ), where ϕT1- angle image of the current frame relative to the image of the first frame, Δα=αTF-αRthat Δβ=βTF-βR. 4. The method according to claim 1, wherein the interframe Phi is Tracey result of threshold processing is done according to the rule
where gn(i, j) is the result of interframe filtering in the form of a binary image;
K - the number of analyzed frames preceding the current;
W is the sliding window size, placed sequentially in each point of the current frame (N=3, 5, 7,...); D - the minimum required number of frames for a decision on the membership of the considered point of the current frame object; n=2, 3, 4,... is the frame number.
|