Online multi-target tracking method and system based on twin network and long-term and short-term clues

Abstract

The invention discloses an online multi-target tracking method and system based on a twin network and long and short-term clues, and belongs to the field of multi-target tracking. The system comprises: a twin network module configured to perform cross-correlation of a tracking target template and a search area to obtain a response graph and obtain a preliminarily predicted tracking trajectory of each target; a correction module which is used for combining the preliminary track and the observation frame together and correcting the pedestrian frame through a pedestrian regression network; a dataassociation module which is used for calculating the similarity between the tracking tracks and the observation pedestrians, respectively extracting long-term and short-term clues of the tracking tracks and the observation pedestrians and fusing the long-term and short-term clues so as to calculate the similarity, and allocating a corresponding observation pedestrian box to each tracking track; and a track post-processing module which is used for updating, supplementing and deleting the tracking track to complete the tracking of the current frame. According to the method, the problems of apparent feature fusion, pedestrian interaction occlusion and large-scale change in a multi-target tracking task are perfected, the accuracy is improved, and the problem of feature misalignment is relieved.

Description

technical field [0001] The invention belongs to the technical field of multi-target tracking, and more specifically relates to an online multi-target tracking method and system based on twin networks and long-term and short-term clues. Background technique [0002] In the face of increasingly complex video scenes, in order to achieve effective processing of massive video data, it is necessary to detect, locate, track and analyze all meaningful objects in the video. Among them, MultiObject Tracking (MultiObject Tracking) / Multi Target Tracking), as a middle-level visual task, plays a key role in it, especially when the new crown pneumonia is sweeping the world in 2020, real-time monitoring of urban communities, road traffic, and accurate tracking of home isolation and crowds through security cameras Pedestrians, outsiders, and vehicles entering and leaving the gathering place have great practical significance for epidemic monitoring. Multi-target tracking is aimed at complex...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the problem of interactive occlusion between tracking targets and the defects and improvement needs of large-scale morphological changes in the existing multi-target tracking task, the present invention provides an online multi-target tracking method and system based on Siamese networks and long-term and short-term cues. The purpose is to maximize the fusion of apparent features and pedestrian interactive occlusion in multi-target tracking tasks, solve the problem of large-scale changes, greatly improve the accuracy and accuracy of data association, and alleviate the problem of feature misalignment

Images