Low-rank redetection context long-time tracking method and system based on residual compensation

Abstract

The invention discloses a low-rank redetection context long-time tracking method and system based on residual compensation. The method is used for tracking a target object appearing in a video, and comprises the following steps: according to the position of the target object in each frame of image in the video, determining a target region to which the target object belongs, extracting at least twoimage features in the target region, and constructing a context feature set; associating the contextual feature set with a corresponding position relation of features in the feature set by a likelihood distribution function to obtain a target object tracking model; and in the process of tracking the target object, using the first frame of image where the target object is as a reference image, calculating tracking residual errors of subsequent frames of images and the reference image one by one, and then compensating to the target object tracking model to predict the position of the target object and store the position. The method improves the accuracy and stability of tracking.

Description

technical field [0001] The invention belongs to the field of long-term context tracking, in particular to a low-rank re-detection context long-term tracking method and system based on residual compensation. Background technique [0002] In the field of computer vision, target tracking technology is widely used in many fields such as visual surveillance, live event broadcasting, unmanned driving, and military navigation. In the past ten years, a large number of excellent visual tracking methods have emerged, but they are affected by factors such as motion blur, illumination changes, noisy backgrounds, occlusions, and scale changes. To study a robustness of the target in complex scenes The method of tracking remains an open question. [0003] Recently, discriminative tracking methods based on correlation filtering have made great achievements. Since Bolme used the proposed least sum of squares (MOSSE) filter for target tracking, due to its computational efficiency and robustn...

AI Technical Summary

Problems solved by technology

[0005] Although the STC algorithm uses context information to achieve good results in tracking, there are some other important flaws that make it difficult to achieve long-term tracking:

[0006] (1) STC uses a single grayscale information feature, and does not make reasonable use of the difference in visual attention in all color channels, and the discriminative feature is not strong;

[0007] (2) Capture the error accumulation caused by the tracking process during tracking, which affects the robustness and accuracy of the long-term tracking algorithm, and ultimately affects the accuracy of target tracking;

[0008] (3) The lack of a stable compensation mechanism for capturing tracking errors makes tracking unable to meet long-term tracking;

[0009] (4) The simple template update strategy and the lack of a re-detection mechanism make the algorithm unable to re-initialize when tracking failures caused by occlusions occur

Images