A multi-target tracking method based on time series multi-feature fusion

Abstract

The invention proposes a multi-target tracking method based on time series multi-feature fusion. The method of the present invention obtains the category of the tracking target and the candidate frame according to the multi-target detection algorithm; uses the convolution network and the correlation filter to calculate the movement prediction center point and screens the candidate frame; calculates the appearance similarity score; calculates the motion similarity score; calculates the interaction feature Similarity score; transform the candidate frame in the tracking frame of the current frame image after screening, update the feature information of the tracking target; calculate the movement prediction center point of the tracking target that does not match the candidate frame and filter the candidate frame; for existing tracking The target correlates the unmatched candidate frames to construct a new tracking target; calculates the overlap between each tracking target by using the cross-over-union ratio; identifies the tracking target that is continuously in the missing state in multiple frames of images as the target that has disappeared. Compared with the prior art, the present invention improves the tracking precision.

Description

technical field [0001] The invention relates to the technical fields of computer vision and target tracking, in particular to a multi-target tracking method based on time series multi-feature fusion. Background technique [0002] Target tracking means that in the image sequence, the target that the system is interested in is detected first, the target is accurately positioned, and then the target's motion information is continuously updated during the moving process of the target, so as to achieve continuous tracking of the target. Target tracking can be divided into multi-target tracking and single-target tracking. Single-target tracking only focuses on one target of interest. Its task is to design a motion model or appearance model to solve the influence of factors such as scale transformation, target occlusion, and illumination, and to calibrate the sense frame by frame. The image location corresponding to the object of interest. Compared with single-object tracking, mul...

AI Technical Summary

Problems solved by technology

[0005] At present, the more popular way in the field of multi-target tracking is to rely on the data association algorithm of the detector. This kind of method solves the problems of target initialization, extinction, and scale transformation well, but it still cannot solve the problem of excessive dependence on the performance of the detector. , Mutual occlusion between multiple targets, similar appearance target distinction, etc.

Images