Target tracking method of sparse cooperation model of hybrid blocks

Abstract

The invention discloses a target tracking method of a sparse cooperation model of hybrid blocks. The method comprises: initializing a target and selecting a target area according to a first-frame image, and carrying out random sampling around the target area through particle filtering, affine transformation and image interpolation to acquire positive and negative sample templates of the target and candidate samples; using a k-d tree to search for the best candidate target as a training dictionary; using a sliding window to partition the dictionary, and using incremental singular value decomposition to obtain a mean value and characteristic values of the dictionary; calculating a confidence value of each candidate target through a sparse expression model and the positive and negative sample templates; using the sliding window to partition the candidate targets, and establishing the sparse expression model; and acquiring the best candidate target of a current frame through posterior probability maximization, and updating the negative sample templates and the dictionary at intervals. The method has the advantage that the tracking precision of the target can be effectively improved through a combination of tracking an overall block and local blocks of the target when the tracked target is in a complex background.

Description

technical field [0001] The invention relates to the technical field of visual tracking, in particular to a target tracking method of a mixed block sparse cooperative model. Background technique [0002] Intelligent video surveillance is an application direction that has developed rapidly in the field of computer vision in recent years and has been studied more. It can use computer vision technology to process, analyze and understand the collected video signals, and based on this Control, so that the video surveillance system has better intelligence and robustness. [0003] In the acquired video sequence, the tracking target may have deformation, illumination change, scale change, occlusion, complex background and so on. In most sparse representation-based tracker algorithms, only the overall representation is considered, and the sparse coefficients are not fully utilized to distinguish the target from the background, and therefore there may be similar objects, partial occlu...

AI Technical Summary

Problems solved by technology

[0003] In the acquired video sequence, the tracking target may undergo deformation, illumination change, scale change, occlusion, complex background, etc.

In most sparse representation-based tracker algorithms, only the overall representation is considered, and the sparse coefficients are not fully utilized to distinguish the target from the background, and therefore there may be similar objects, partial occlusions, fast movements, etc.; therefore, there is complexity in tracking targets Tracking is prone to failure when the background

[0004] In the existing algorithm, the problem of similar objects or occlusion of the tracked target during the tracking process is better solved, but there are still tracking targets that fail to be tracked, and there is a big gap between the real-time performance and the actual system requirements. Therefore, improving the real-time performance of the target tracking algorithm is still a very challenging subject.

Images