Online multi-pedestrian detection tracking method in complex scene

Abstract

The invention provides an online multi-pedestrian tracking algorithm research based on detection. A target detection network YOLOv3 based on deep learning is used as a detector; a deep learning network is used to extract pedestrian features and Kalman filtering is used to predict pedestrian motion position information. A joint measurement mode based on detection confidence, apparent similarity andmotion similarity is provided to measure correlation between detection and tracking, an algorithm for adaptively adjusting weight factors of the apparent similarity and the motion similarity is provided, and finally, a KM matching algorithm and an IOU matching algorithm are adopted to realize real-time matching of detected pedestrians. Experimental results show that online multi-pedestrian detection and tracking can be realized in a complex scene, and high accuracy is realized. The practical application value of online multi-pedestrian detection and tracking is particularly outstanding, and the method is widely applied to the fields of intelligent video monitoring, automatic driving, robot vision navigation, human-computer interaction and the like.

Description

technical field [0001] The invention relates to online pedestrian detection and online pedestrian tracking in computer vision, especially the online pedestrian detection and tracking in complex scenes, that is, the real-time acquisition of position coordinate information and motion tracks of pedestrians in video. Background technique [0002] As a key technology in computer vision, multi-target detection and tracking has attracted more and more attention. Among them, the practical application value of multi-pedestrian detection and tracking is particularly prominent. It is widely used in intelligent video surveillance, automatic driving, robot visual navigation, Human-computer interaction and other fields. [0003] Target tracking algorithms are mainly classified into single target tracking and multi-target tracking. Compared with multi-target tracking algorithms, the research on visual single target tracking algorithms is more extensive and mature. Typical ones include Mean...

AI Technical Summary

Problems solved by technology

However, the multi-objective algorithm is still in the exploratory stage, and there is no better solution. The mainstream research is to optimize and improve on the detection-based tracking framework.

Images