Vehicle multi-target detection and trajectory tracking method based on re-identification

Abstract

The invention discloses a vehicle multi-target detection and trajectory tracking method based on re-identification, and belongs to the field of computer vision technology and video monitoring. The method comprises the following steps that firstly, a model capable of effectively extracting vehicle visual features on each track is trained by utilizing a vehicle re-identification technology; then, vehicle detection is carried out on each frame of image in the camera; and finally, in combination with the vehicle appearance features extracted by the re-identification model, motion prediction and visual similarity discrimination are fused, and single-camera multi-target tracking is executed, so that the driving trajectories of all matched vehicles are obtained. According to the method, multi-target matching of front and rear frames of images is successfully achieved in the aspect of tracking, the problem of tracking target shielding is effectively solved, and the appearance characteristics of the vehicle images shot by the multi-angle camera are extracted by adopting a re-identification technology aiming at the possible instability of algorithm detection performance caused by multiple visual angles and are fused into a tracking scheme, re-identification which occurs again after the tracking target is lost is achieved, and the track tracking capability is further improved.

Description

technical field [0001] The invention relates to the fields of computer vision technology and video monitoring. Background technique [0002] In recent years, research related to vision is developing rapidly all over the world. In many literatures, the two topics of computer vision and machine vision are considered as one, but in fact the two are both different and related. Machine vision mainly focuses on the analysis of quantities, such as the visual measurement of railway turnout gaps. The focus of computer vision research is to assist in the perception of three-dimensional scenes and the understanding and understanding of the objective world. It usually uses a combination of multiple intelligent techniques, and utilizes complex knowledge representations, using heuristic search and matching techniques to control strategies. Computer vision emphasizes the research on theoretical algorithms, mainly qualitative analysis, such as behavioral analysis of semantic segmentation...

AI Technical Summary

Problems solved by technology

Although this method of applying traditional machine learning to vehicle detection and tracking has certain advantages, many of these classifiers are still weak classifiers, the robustness of the algorithm is insufficient, and there is often a risk of high false alarm rate

[0010] Furthermore, many existing patented tracking solutions mainly focus on the single target tracking problem, so if the multi-target tracking requirements need to be achieved by tracking the detected targets one by one

Images