Online multi-target tracking algorithm fusing single-target tracking result

Abstract

The invention discloses an online multi-target tracking algorithm fusing a single-target tracking result. The algorithm comprises the following steps: predicting the current position of a target by utilizing a single-target tracking algorithm CSRT; utilizing a ReID network model to respectively extract feature vectors from the prediction frame area and the detection frame area, and calculating thesimilarity of the feature vectors; constructing a weighted undirected graph in combination with the target, the prediction box and the detection box; performing clustering operation on the undirectedgraph by utilizing a Markov clustering algorithm to obtain a cluster; and finally, updating the current position of the target according to the prediction box and the detection box in the cluster. According to the online multi-target tracking algorithm, the multi-target tracking precision is improved, and the situation of missing detection of the detection frame due to target shielding, small target size and the like is effectively improved.

Description

technical field [0001] The invention relates to the field of visual tracking, in particular to an online multi-target tracking algorithm for fusing single target tracking results. Background technique [0002] Multiple object tracking in video is a fundamental and important task for many vision applications, such as video surveillance and autonomous driving. The goal of this task is to localize multiple objects in each frame and obtain trajectories for each identity. At present, most methods are based on the detection frame tracking method, including online tracking and offline tracking: the online tracking method is to construct an association matrix according to the similarity between the target and the detection frame, and use the matching algorithm to match the position of the target and the detection frame; offline tracking The method is to construct a graph based on the detection frames in a period of time and the similarity between them, and use subgraph division to ...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to provide an online multi-target tracking algorithm that combines single target tracking results, and uses Markov clustering algorithm to cluster the targets, prediction frames, and detection frames of two frames in combination with the prediction results of a single target tracker , use the prediction frame to solve the problem of missed detection by the detector due to target occlusion or too small target

Images