A target tracking method and system based on target detection

Abstract

The invention discloses a target tracking method and system based on target detection, in the technical field of pattern recognition, including: establishing a target tracking model including a backbone network and a reasoning network for the target to be tracked, and the backbone network is a trained target detection model Feature extraction network; in the first frame F in the target video 1 Create multiple positive samples above, initialize the inference network through model training; sequentially obtain the frames in the target video as the current frame, use the target tracking model to obtain multiple candidate regions in the current frame and their detection scores and target position coordinates; according to the spatial constraints Correct the detection score to obtain the tracking score; if the tracking score of the target area with the highest tracking score is Score>Th S , then the tracking is successful, and positive and negative samples are created and added to the sample library; if Th F ≤Score≤Th S , it is determined that the tracking is successful; if the Score <thF, then it is judged that the tracking fails, and the inference network in the sample library is used for online update. The invention can realize the continuous and accurate tracking of the target.< / th

Description

technical field [0001] The invention belongs to the technical field of pattern recognition, and more specifically relates to a target tracking method and system based on target detection. Background technique [0002] Target detection, also called target extraction, is an image segmentation based on target geometric and statistical features. It combines target segmentation and recognition, and its accuracy and real-time performance are an important capability of the entire system; target tracking , that is, given the size and position of the target in the initial frame of a video sequence, predict the size and position of the target in subsequent frames. Due to the strong discrimination and regression capabilities of deep learning-based target detection algorithms, some techniques in target detection algorithms have received more and more attention and applications in deep learning-based tracking algorithms. [0003] However, it is infeasible to directly apply deep learning...

AI Technical Summary

Problems solved by technology

Single target tracking is a class-free task, that is, the specific category and shape of the target are not known before online tracking, which makes it difficult to find effective training samples during offline training

Second, the single target tracking task has complex motion properties

[0004] In general, due to the above-mentioned differences between the target tracking task and the target detection task, the existing target tracking methods that apply target detection technology cannot make good use of the discrimination and regression capabilities of the target detection algorithm, and the tracking effect is poor. , it is very likely to lose the target during the target tracking process

Images