Incremental Subspace Target Tracking Method Based on lp Norm Regularization

Abstract

The invention discloses a p The incremental subspace target tracking method with norm regularization includes the following steps: read in the video sequence image, manually mark the first frame target image; use the Ross method to obtain the observation value; directly calculate the residual error for the first 5 frame images to Determine the target state of the current frame, collect samples, and use Ross’s method to obtain the initial subspace and mean vector; starting from the sixth frame, establish an objective function for the observation samples under the framework of augmented Lagrangian, and perform Minimize the calculation; evaluate the confidence of each candidate sample as the real target of the current frame, and collect samples; when the collected samples reach 5, use Ross's method to incrementally update the subspace and the mean vector. Under the premise of ensuring a certain real-time performance, the present invention has strong anti-interference performance for targets under challenges such as occlusion, illumination, and motion blur, and can observe the influence of different sparsity on tracking under a unified framework .

Description

Technical field: [0001] The invention belongs to the field of machine vision, in particular to a p Norm-Regularized Incremental Subspace Object Tracking Approach. Background technique: [0002] Object tracking in video has important applications in video surveillance, human-computer interaction, behavior analysis and other fields. Although in the past ten years, experts and scholars from various countries have proposed many methods and made a lot of exciting progress in this field, it is still a very challenging task to track the target in real time and robustly. One of the reasons is that the video is a dynamic time sequence, and the posture of the target will change during the travel process, and may encounter interference from lighting, occlusion, and similar objects. At the same time, in the process of traveling, the shaking of the camera or the rapid movement of the target will also cause the appearance of the target to be blurred, which further aggravates the difficu...

AI Technical Summary

Problems solved by technology

[0004] However, the method based on subspace collaborative representation needs to make full use of each feature component in the subspace to reconstruct the target, and inevitably introduce redundant feature components

In addition, the processing of occlusion is still inappropriate. For example, the rectangular template in [2] has the ability to reconstruct the foreground and background at the same time. When each rectangular template is used for collaborative representation, it is bound to further deteriorate the two. meaning, causing the target to deviate or even lose

Images