Compression tracking method combining local structured information

Abstract

The invention provides a compression tracking method combining local structured information. In compression tracking, an information structured preprocessing step is added. Adjacent pixels with a similar texture, color, brightness and other characteristics are aggregated into an irregular pixel block possessing a visual sense meaning. Based on the pixel block, an HSV color characteristic is extracted. Through constructing a color confidence graph, the rough estimation of one phase of a target position is performed. And through combining the local structured information random projection of thepixel block and the pixel-level random projection, a novel measurement matrix is formed. A measurement matrix is used to carry out dimension reduction processing on a haar-like characteristic. Two-phase accurate position estimation of a target is completed according to a Naive Bayes classifier so as to achieve accurate target tracking. Compared with an existing tracking technology, by using the method of the invention, through combining the local structured information of a middle-level clue and the haar-like characteristic of a low-level clue, a target appearance is represented, the accuracyand the robustness of target tracking are effectively increased and the real-time performance of the target tracking is guaranteed.

Description

technical field [0001] The invention relates to a compression tracking technology, in particular to a compression tracking method combined with local structured information, which belongs to the video analysis technology in the field of video data processing or generation, and belongs to the field of video analysis technology. Background technique [0002] With the application of video surveillance technology in the field of security more and more widely, target tracking plays an important role in computer vision systems, and has broad prospects in many fields such as intelligent transportation and security monitoring. In recent years, target tracking technology has also been a hot research topic for many scholars. However, due to the complexity of targets and backgrounds, such as illumination changes, target occlusion, background blur, etc., how to update the appearance model correctly and in real time, and estimate accurately and efficiently The location of the target is s...

AI Technical Summary

Problems solved by technology

This method applies multi-feature fusion to compressed tracking, which improves the accuracy of tracking, but the computational complexity it brings is more than doubled.

[0005] The above-mentioned methods have improved the accuracy of tracking, but still stay at the image pixel level for processing, lack of higher-level feature utilization, and the tracking efficiency is still not ideal. Yang introduced superpixels into tracking in 2014 and proposed Robust Superpixel Tracking algorithm (Robust Superpixel Tracking), which uses superpixels to maintain most of the boundaries of the target object in the image, and contains more local information than pixels, transforms the tracking problem into a computational goal- Background Confidence Map Problems for Accurate and Robust Tracking

However, the algorithm uses meanshift clustering, the entire frame structure is lengthy, and the tracking takes too long to meet the real-time requirements

Images