An occlusion detection method based on block correlation

Abstract

The invention discloses an occlusion detection method based on block correlation. A target area is selected, and the target area is sequentially divided into a first number of first target blocks in afirst direction, and the target area is sequentially divided into a second number of second target blocks in a second direction. The method comprises: calculating a first target block of a predetermined position and a normalized cross correlation value of the second target block and the block of a position corresponding to the predetermined template; judging the occlusion degree and the occlusiondirection of the target region according to the normalized cross correlation value, wherein two adjacent first target blocks have a first overlap portion and two adjacent second target blocks have asecond overlap portion. The embodiment of the invention can quickly and effectively judge the process of the target from entering the occlusion to the complete occlusion and then to the outgoing occlusion by dividing the target area into blocks in different directions, and then judging the occlusion degree and the occlusion direction, and carries out corresponding processing.

Description

technical field [0001] The invention belongs to the field of target tracking, and in particular relates to an occlusion detection method based on block correlation. Background technique [0002] In the field of computer vision moving target tracking, the tracked target may be partially or completely occluded by other objects during the movement process. How to solve the occlusion problem has always been one of the research hotspots in the field of video tracking. Traditional target tracking methods such as frame difference method, background difference method, and optical flow method cannot obtain effective information when the target is occluded, resulting in tracking failure. The tracking-by-detection methods that have been popular in recent years are also difficult. The occluded target is detected, and the CF class method of the big fire cannot solve the occlusion problem due to its algorithm nature and framework limitations. [0003] For example, the existing technology...

AI Technical Summary

Problems solved by technology

Traditional target tracking methods such as frame difference method, background difference method, and optical flow method cannot obtain effective information when the target is occluded, resulting in tracking failure. The tracking-by-detection methods that have been popular in recent years are also difficult. The occluded target is detected, and the CF method of the big fire cannot solve the occlusion problem due to its algorithm nature and framework limitations.

[0003] For example, the existing technology provides a multi-block tracking algorithm based on Mean Shift. The occlusion processing method of this algorithm performs Mean Shift iterations on each block, which has a large amount of calculation and poor real-time performance. At the same time, only external block information is used for occlusion judgment. It is not ruled out that the peripheral block contains a large amount of background information, which is easy to be misjudged as occlusion due to the change of the background, and cannot accurately determine the detailed occlusion of the target; The accurate positioning after the emergence is not considered, which may easily lead to the loss of the target; the existing technology also provides a multi-block tracking algorithm based on the spatio-temporal context visual tracking algorithm (STC). Multiple tracking points within the target are tracked by the STC algorithm. First, evenly take multiple tracking points on the target, and use the STC algorithm to track a single tracking point to obtain the position of each tracking point in the current frame.

When the target is partially occluded, the position of the tracking point will be abnormal. At this time, most points with similar trends are taken as reliable points. Through the unoccluded area and effective background area corresponding to the tracking point, the target can still be found, and the tracking effect is good. Therefore, the algorithm has a strong ability to resist local occlusion. However, the above tracking algorithm uniformly collects points for the target, and uses the STC algorithm to track multiple tracking points, which takes a long time, and these tracking points do not have strong characteristics. It is very likely that the abnormality of the tracking point is caused by the deformation and rotation of the target, and then it is misjudged as occlusion. At the same time, the effective tracking point cannot be found after the target is completely occluded, resulting in tracking failure.

[0004] To sum up, none of the existing schemes can effectively judge the occlusion situation, and it is easy to cause the target to be lost when the target encounters occlusion, resulting in tracking failure, and the calculation process is complicated

Images