Multi-target tracking method combined with spatiotemporal motion

Abstract

With the increasing amount of video data, the use of computer vision technology to extract effective information from video is an urgent task, and multi-target tracking is an important part of it. To this end, the present invention proposes a multi-target tracking method combining spatiotemporal motion. Firstly, a spatiotemporal motion model based on the relative position relationship between targets is proposed, which uses the relative motion between the targets to estimate the position of the target robustly, and uses the GRU model to fit the relative motion between the targets to improve the accuracy of the final position prediction. Then, a multi-stage target association algorithm based on the historical tracking state of the target is proposed. First, the target with better historical association results is used as the anchor point, then the position of the remaining targets is estimated based on the anchor point position and the relative motion model, and finally the remaining targets are used. The associations are performed in sequence according to different tracking states. Finally, the tracking results on the test data show that the multi-target tracking algorithm proposed by the present invention has obvious advantages.

Description

technical field [0001] The invention relates to a multi-target tracking method, in particular to a multi-target tracking method combined with space-time motion, belonging to the field of computer vision. Background technique [0002] With the development of society and economy, the cost of hardware such as cameras, mobile phones and surveillance cameras has gradually decreased, and Internet technology has continued to improve. A large amount of video data is produced every day for security monitoring, driverless driving, and so on. These massive data often contain important information, and it is time-consuming and laborious to distinguish and extract the information only manually. Therefore, it is very meaningful and urgent to hand over the data to the machine for processing. Existing computer vision tasks mainly include object detection, object classification and recognition, object tracking, and scene semantic analysis. [0003] Object tracking is an important part of co...

AI Technical Summary

Problems solved by technology

In order to solve the problem of large deviation in the position estimation of the single target motion model when the moving camera or the camera is in large motion, the present invention implements a spatiotemporal motion model based on the relative positional relationship between targets on the basis of the single target time series motion model

Images