Research on a multi-target detection and tracking algorithm

Abstract

Aiming at the problems of poor real-time performance and easy drift in the existing research methods in the field of multi-target tracking,a multi-target detection and tracking algorithm was proposed based on YOLOv3 algorithm and KCF algorithm.Firstly,the trained YOLOv3 network was used to obtain the location of the target in the video,and the ID of each target was allocated;Secondly,multiple targets were input into the tracking module based on kernel correlation filter in parallel for target tracking;Then,the conditions for starting the correction strategy were judged,if they were met,the results of the detection module were used to correct the results of the tracking module;Finally,the kernel correlation filter model was updated by using the tracking results.The experimental results show that when the algorithm is applied to four groups of video sequences containing multiple interferences in OTB2015 data set,the tracking accuracy reaches 82.4%,the tracking success rate reaches 81.1%,and meets the requirements of real-time tracking.Therefore,the algorithm is not only valid,but also has stronger robustness to provide a new research method for the field of multi-target tracking.

Keywords

• computer neural network; multi-target detection and tracking; yolov3; kerneized correlation filter algorithm; correction strategy