Method for selection of low-contrast objects from images

Abstract

FIELD: information and measurement technologies.SUBSTANCE: invention can be used to detect and isolate the contours of moving objects from images obtained by high-speed optical channels in the visible range. The video stream received by the digital video stream sensor in real time is divided into a set of successive halftone images of dimension M×N pixels, represented as two-dimensional data arrays with brightness values on a scale from 0 to 255, converted into two information formats. Next, the temporal difference of adjacent frames in two information formats is processed in parallel and distributed into two streams, the first of which, in turn, is processed in an averaging spatial filter with a neighbourhood size of L=5 and segmented according to the criterion of the maximum brightness value, the second stream, in turn, is processed in parallel in an averaging spatial filter with a neighbourhood size of L=9, a spatial unsharp masking filter with a neighbourhood of L=5, an adaptive filter with a neighbourhood of L=5 with subsequent segmentation according to the criterion of the current brightness exceeding threshold values, which are subject to combining the results of segmentation of streams of difference images in the processed information formats to determine common areas corresponding to the object in the images.EFFECT: increase in the accuracy and reliability of detecting objects and determining their movement parameters.1 cl, 1 dwg

Description

[0001] The proposed invention relates to the field of information and measurement technologies and can be used to detect and highlight the contours of moving objects from images obtained by high-speed optical channels in the visible range.

[0002] A method is known for processing a sequence of images for detecting and tracking aerial objects [RU 2419150, 2010], which consists in sequentially performing temporal filtering of the observed image formed by television and / or infrared range sensors with an exponential filter, calculating the difference between the observed image and the obtained time-smoothed image multiplied by a weighting coefficient, parallel two-dimensional averaging spatial filtering of the temporal difference with two filters with masks of different types and dimensions, obtaining the difference in the results of spatial filtering, finding an estimate of the standard deviation of noise, calculated over the entire area of ​​the frame, which is used in the operation of threshold processing of the difference in the results of spatial filtering with closing and opening with a square structuring element, performing a procedure for marking and parameterizing a binary image,obtained after morphological operations, lists of parameters of connected areas of the image are obtained, according to which a decision is made about the presence of objects in the processed area of ​​the image frame, after which a tracking strobe is formed for each detected object depending on its size and speed of movement, followed by selection of objects in the strobe.

[0003] A known method for processing a sequence of images against the background of the starry sky [RU 2624828, 2017] consists of receiving and analog-to-digital converting the image signal of each frame and in detecting an aircraft and estimating its trajectory parameters, by combining image processing algorithms, including an object detection algorithm, a small object tracking algorithm, a large object tracking algorithm, and a cone tracking algorithm, switching between which is carried out depending on the values ​​of the aircraft size estimates and the distance to it; detecting the aircraft and estimating its trajectory parameters using a selected algorithm, while switching between the detection, large object tracking, and cone tracking algorithms is performed as follows: first, the object is detected using the detection algorithm;After the object is detected, a switch is made to the small-sized object tracking algorithm; when the object reaches a large size, the large-sized object tracking algorithm is used; if the object is lost during the operation of the object tracking algorithms, the object is detected again by the detection algorithm; when a specified distance to the object is reached, at which the cone image and the known reference image correspond in size, a search for the cone occurs and a switch is made to the cone tracking algorithm, which terminates when the boom of the refueled aircraft hits the object's cone.

[0004] The closest in technical essence to the proposed method is the method of selecting images of an object on a non-uniform background [RU 2734655, 2020], which consists of forming a compensating signal by smoothing the current digital signal in an aperture with a size m>M o , but m<<M ф , where Mo and M ф - the sizes of the image of the object and the background, respectively, the formation of a difference signal by subtracting the compensating signal from the current one and the formation of an output binary signal of the object by comparing the difference signal with a fixed threshold, characterized in that in each i-th frame of the video sequence, an estimate of the size of the binary image of the object M is made oi , at l <M oi ≤m perform preliminary smoothing of the current signal with aperture m i =M oi , and when the condition w is reached i =m the aperture size is fixed.

[0005] The disadvantages of the prototype are:

[0006] - low selection accuracy for high-speed objects of observation and the appearance of noise of various natures;

[0007] - dependence of the accuracy of object selection on the quality of the background model construction;

[0008] - low efficiency of the method in the presence of a moving background.

[0009] The objective of the proposed method for selecting low-contrast objects from images is to increase the accuracy and reliability of detecting objects and determining their motion parameters.

[0010] The proposed method for selecting low-contrast objects from images allows for high selection accuracy under conditions of camera movement, background changes, for high-speed objects of observation, and the imposition of additive and impulse noise.

[0011] The essence of the proposed method for low-contrast objects in images is to form a compensating signal by smoothing the current digital signal in an aperture with a size m>M o , but m<<M ф , where M o and M ф- the sizes of the image of the object and the background, respectively, the formation of a difference signal by subtracting the compensating signal from the current one and the formation of the output binary signal of the object by comparing the difference signal with a fixed threshold, performing an estimate of the size of the binary image of the object M oi in each i-th frame of the video sequence, preliminary smoothing of the current signal with aperture m i =M oi at l <M oi ≤m, fixing the aperture size when the condition m is reached i=m, characterized in that the temporal difference of adjacent frames is processed in parallel in two information formats and distributed into two streams, the first of which, in turn, is processed in an averaging spatial filter with a neighborhood size of L=5 and segmented according to the criterion of the maximum brightness value, the second stream, in turn, is processed in parallel in an averaging spatial filter with a neighborhood size of L=9, a spatial unsharp masking filter with a neighborhood of L=5, an adaptive filter with a neighborhood of L=5 with subsequent segmentation according to the criterion of the current brightness exceeding threshold values, which are subject to combining the results of segmentation of streams of difference images in the processed information formats to determine common areas corresponding to the object in the images.

[0012] The system implementing the proposed method is shown in Fig. by the following positions:

[0013] 1 - digital video stream sensor;

[0014] 2 - uint8 data format converter;

[0015] 3 - double data format converter;

[0016] 4 - time difference generation device;

[0017] 5 - time difference generation device;

[0018] 6 - ​​random access memory of the previous frame;

[0019] 7 - random access memory of the previous frame;

[0020] 8 - averaging spatial filter with neighborhood L=5;

[0021] 9 - averaging spatial filter with neighborhood L=9;

[0022] 10 - soft masking filter with high frequency boost;

[0023] 11 - adaptive filter;

[0024] 12 - threshold segmentation device;

[0025] 13 - threshold segmentation device;

[0026] 14 - object detector.

[0027] The output of the digital video stream sensor 1 is connected to the input of the data format converter 2 and the data format converter 3, the output of the data format converter is connected to the input of the time difference generating device 4, the output of the time difference generating device 4 is connected to the input of the operational memory of the previous frame 6 and the averaging spatial filter with the neighborhood 8, the output of the averaging spatial filter 8 is connected to the input of the threshold segmentation device 12, the output of the data format converter 3 is connected to the input of the time difference generating device 5, the output of the time difference generating device 5 is connected to the input of the operational memory of the previous frame 7 is connected to the inputs of the averaging spatial filter 9, the unsharp masking filter with high-pass lift 10 and the adaptive filter 11, the outputs of which are connected to the threshold segmentation device 13,the output of the threshold segmentation device 12 and the output of the threshold segmentation device 13 are connected to the object detector 14.,

[0028] The method is as follows.

[0029] The video stream received by digital video stream sensor 1 in real time is broken down into a set of successive halftone images with dimensions of M×N pixels, represented as two-dimensional data arrays with brightness values ​​on a scale from 0 to 255 and converted in data format converter 2 to the uint8 information format and data format converter 3 to the double information format. In time difference generation devices 4 and 5, time differences are simultaneously generated by pixel-by-pixel subtraction of the brightness values ​​of the current frame and the previous frame in the video stream, stored in the previous frame random access memories 6 and 7, taking into account the frame shift by rows and columns.Compensating signals are generated in parallel in spatial averaging filters 8 and 9 for both information formats to smooth out grouped difference pixels corresponding to the appearance of false object location points by convolving the temporal difference with the filter mask of the desired neighborhood. Simultaneously, the contrast of the temporal difference is increased in unsharp masking filter 10 by introducing high-frequency components. At the same time, excessive thinning or thickening of object boundaries is smoothed in adaptive filter 11. Next, threshold segmentation device 12 determines the membership of a temporal difference element based on the criterion of the current brightness exceeding the set threshold values. Simultaneously, threshold segmentation device 13 determines the membership of a temporal difference element based on the criterion of the current brightness exceeding the set threshold values ​​and logically combines the filtering results.The final result of object detection is formed in detector 14 by determining the general areas corresponding to the object.

[0030] Thus, the proposed method for selecting low-contrast objects from images will improve the accuracy and reliability of detecting objects and determining their motion parameters through parallel processing of time differences, taking into account:

[0031] - smoothing of grouped difference pixels during spatial filtering;

[0032] - adding high-frequency components that enhance the contour of the object;

[0033] - smoothing out excessive thinning or thickening of the boundaries of objects;

[0034] - definitions of common areas corresponding to an object based on the results of processing time differences in two information formats.

Claims

A method for selecting low-contrast objects from images, which consists of forming a compensating signal by smoothing the current digital signal in an aperture with a size m>M o , but m<<M ф , where M o and M ф - the sizes of the image of the object and the background, respectively, the formation of a difference signal by subtracting the compensating signal from the current one and the formation of an output binary signal of the object by comparing the difference signal with a fixed threshold, performing an estimate of the size of the binary image of the object M oi in each i-th frame of the video sequence, preliminary smoothing of the current signal with aperture m i =M oi at 1 <M oi ≤m, fixing the aperture size when the condition m is reached i=m, characterized in that the temporal difference of adjacent frames is processed in parallel in two information formats and distributed into two streams, the first of which, in turn, is processed in an averaging spatial filter with a neighborhood size of L=5 and segmented according to the criterion of the maximum brightness value, the second stream, in turn, is processed in parallel in an averaging spatial filter with a neighborhood size of L=9, a spatial unsharp masking filter with a neighborhood of L=5, an adaptive filter with a neighborhood of L=5 with subsequent segmentation according to the criterion of the current brightness exceeding threshold values, which are subject to combining the results of segmentation of streams of difference images in the processed information formats to determine common areas corresponding to the object in the images.

Images