RU2557484C1 - Image segmentation method - Google Patents

Abstract

FIELD: physics.

SUBSTANCE: to detect parts of an image containing moving objects, the method includes nulling pixels with the same numbers in both frames consecutively from the edges of the image to the centre; for each nulled pixel, measuring an addition to the interframe difference of frequency characteristics at low phase numbers p; determining the number of the nulled pixels, wherein said addition exceeds a set threshold; based on the values of said numbers of pixels, selecting a segment of the image containing a moving object.

EFFECT: faster segmentation.

5 dwg

Description

The invention relates to the field of image segmentation, and more particularly to the allocation of image areas containing moving objects.

A known method for the analysis of inter-frame difference (K. Fu, R. Gonzalez, K. Lee. Robotics / M .: Mir, 1989, p. 423-426). It consists in determining the changes between two image frames, one of which having stationary components, contains a reference object, and subsequent threshold processing of the resulting difference, as a result of which the image pixels of a moving object are extracted. Using the received segmented image, you can determine the location and motion parameters of the object.

The disadvantages of this method are: the difficulty of detecting the movement of an object and determining the parameters of its movement in cases of changes in illumination within the boundaries of the object, a high noise level, especially at low light, with small size or low contrast of a moving object.

There is a method of determining the movement of an object based on measuring interframe differences in frequency characteristics (Bogoslovsky A.V., Zhigulina I.V., Kopylov O.E., Yakovlev V.A. Determining the parameters of movement of an object from an image based on interframe differences in frequency characteristics // Radio Engineering, 2010, No. 5). It consists in measuring the frequency characteristics of the image in rows and columns and further determining their differences. By the differences obtained, it is possible to detect the movement of an object. It involves the automatic determination of the movement of an object relative to a fixed background with good lighting and a high level of contrast of a moving object (low noise).

The disadvantage of this method is the low probability of automatically detecting the movement of the object in cases of high noise, for example, in low light, small size or low contrast of a moving object.

Closest to the technical nature of the claimed method is the "Method of detecting moving objects." Patent RU No. 2461067, 2011, G06T 7/20. It consists in measuring the frequency characteristics of the image in rows and columns and further determining the extreme values of the double differences of the frequency characteristics and comparing them with those obtained previously.

The disadvantage of this method is the need to perform a large number of actions, which reduces the speed of the system as a whole.

The essence of the proposed method is as follows:

detection of moving objects by video sequence is carried out by means of changes in the interframe differences in frequency characteristics that occur when the pixels of the same name are zeroed in two frames of the video sequence. Zeroing is performed sequentially from the edges to the center of the image. When a nullified pixel hits the image of an object in one of two frames, and in the other into the background, the values of additives are close to extreme values. This can occur when the values of the numbers of p phases do not exceed the amount of movement of the object in pixels between frames. Therefore, the numbers of zeroed pixels allow you to determine the edges of a moving object, i.e. segment the image.

The technical result of the proposed method is to increase the speed of automatic detection of moving objects and determine the parameters of their motion by taking into account the connectivity of a moving object.

The specified technical result is achieved by the fact that, to select portions of the image containing moving objects, the pixels with the same numbers are zeroed in both frames sequentially from the edges of the image to the center, for each zeroed pixel, the addition to the interframe difference of the frequency characteristics at small numbers of p phases is measured, determine numbers of zeroed pixels at which the additive in absolute value exceeds the set threshold, an image segment is allocated according to the values of these pixel numbers, neighing moving object; after that, the segmentation process is started again for an image that does not contain a selected area.

The method can be implemented by the following sequence of actions:

- two frames of the video sequence are recorded in memory, and rows (columns) are sampled (see A. Andreev, Automated television surveillance systems. Part 1 / SPb .: SPbGUITMO, 2005, pp. 14-17);

- in the rows (columns) of the first and second frames, pixels with the same numbers are zeroed, zeroing is done sequentially, starting from the edges of the rows (columns) to their center;

- for each zeroing, the additive to the interframe frequency differences is measured at small values of the numbers of p phases;

- values of additives are remembered and compared with positive and negative thresholds;

- the numbers of zeroed pixels are determined at which the values of additives in absolute value exceed the established thresholds;

- the image area containing the moving object is segmented by certain pixel numbers.

The invention is illustrated by drawings, where:

Figure 1 - the proposed structural diagram of a device implementing this method:

1 - shaper video signal;

2 - memory block;

3 - block sampling and zeroing;

4 - block definition of additives;

5 - analysis unit;

6 - block storage segmented image.

Figure 2 - video signal of the same name in the test image of two frames containing a moving object:

- the line contains 101 pixels;

- the entire dynamic range is 255 units .;

- background level a = 130 units;

- the contrast of the object is positive and is b-a = 20 units .;

- the object moves to the right and consists of 20 pixels;

- the left edge of the object in the first frame corresponds to a pixel with the number l = -36, and the left edge of the object in the second frame l + i = -26;

- the right edge of the object, respectively, r = -17 and r + i = -7.

для фазы p=1 при последовательном обнулении одноименных пикселей обеих строк, штриховой линией показаны пороговые значения.Figure 3 is a view of the additives of energy interframe frequency differences $$\Delta S_{off}^j(p)$$

for the phase p = 1, with successive zeroing of the pixels of the same name on both lines, the dashed line shows the threshold values.

Figure 4 - video signals corresponding to figure 2, distorted by additive Gaussian noise with zero mathematical expectation and standard deviation σ = 7.

для фазы p=1 при видеосигнале, соответствующем фиг.4, штриховой линией показаны пороговые значения.5 is a view of the additives of energy interframe frequency differences $$\Delta S_{off}^j(p)$$

for the phase p = 1 for the video signal corresponding to figure 4, the dashed line shows the threshold values.

The implementation of the invention can be illustrated by the example of processing two lines of the same name image frames containing a moving object. As a model of the situation, figure 2 shows the corresponding video signals. When sequentially zeroing the pixels of the same name on both lines, the interframe differences in the frequency characteristics acquire additives, the form of which is shown in Fig. 3 for the phase p = 1. The graph contains two pronounced areas:

- negative additives, they correspond to the numbers of zeroed pixels from -37 to -26, i.e. from l-1 to l + i;

- positive additives for numbers of zeroed pixels from -17 to -6, i.e. from r to r + i + 1.

The extreme values of additives are ± 2Δ (a + b).

As the phase number p increases to (i-1) / 2, the regions where the additives are not equal to zero will expand, while the magnitude of the extreme values does not change. With a further increase in p, the extreme values decrease.

For any movements and signs of contrast, non-zero additives will correspond to the marked numbers of zeroed pixels.

Thus, by measuring the addition of inter-frame differences in frequency characteristics, it is possible to determine the pixel numbers corresponding to the edges of a moving object.

In reality, the background is uneven, it can vary between frames, as well as the video signal corresponding to a moving object, in addition, noise distortions are possible. This forces the introduction of a bilateral threshold and to allocate only those additives that in absolute value will exceed these thresholds.

Figure 4 shows the video signals corresponding to figure 2, distorted by additive Gaussian noise with zero mathematical expectation and standard deviation σ = 7. Figure 5 shows the corresponding additives. The threshold value is selected ± 2Δa = ± 0.26 · 10 ⁴ . It can be seen that even with a sufficiently high noise level, it is possible to distinguish regions corresponding to a moving object. The threshold value depends on the application and can either be set in advance or determined by averaging the signal corresponding to the background over a pre-segmented area. If the background signal corresponds to the middle of the dynamic range, then it is recommended to choose ± (40 ÷ 50) a as a threshold value.

After determining the pixel numbers corresponding to the moving object in both frames, this area containing the moving object is cut off and the segmentation process is repeated again for the remaining part of the image.

Claims (1)

The method of image segmentation, based on the formation of the video signal of the image, storing the sequence of two frames, forcing the image pixels to zero, measuring the interframe frequency differences obtained after zeroing the frames, characterized in that to highlight portions of the image containing moving objects, zeroing the pixels with the same numbers in both frames sequentially from the edges of the image to the center, for each zeroed pixel, the additive to the interframe difference is often measured distinct characteristics at small numbers of p phases, the numbers of zeroed pixels are determined at which the additive exceeds the set threshold in absolute value, the image segment containing a moving object is distinguished from the values of these pixel numbers; after that, the segmentation process is started again for an image that does not contain a selected area.

Images