Self-query-based multi-target tracker and tracking method for unmanned vehicle

Abstract

The invention discloses a self-query-based multi-target tracker and tracking method for an unmanned vehicle, and the method comprises the steps of carrying out the feature information extraction of a video frame photographed by a camera through a ResNet-50 network, and inputting the features into a detection branch and a self-query branch; and performing intersection-parallel ratio calculation on the detection result of the detection branch and the result of the self-query branch to complete tracking. The invention provides a detection branch based on heat map response, the post-processing flow of the detection branch is greatly simplified, and the multi-target tracking speed of the intelligent vehicle is effectively improved; a self-query branch is added, so that the multi-target tracking precision of the intelligent vehicle is improved; the association of object tracks can be completed only by using intersection-to-union matching, so that the post-processing flow of the whole model is reduced, and the real-time performance of multi-target tracking of the intelligent vehicle is effectively improved; the structure of the tracker designed by the invention is modularized construction, the number of corresponding modules can be modified according to different working conditions in actual use, and the reusability and maintainability of the algorithm are ensured.

Description

technical field [0001] The invention belongs to the field of intelligent vehicle vision technology, and particularly designs a self-inquiry-based multi-target tracker and a tracking method for unmanned vehicles. Background technique [0002] Smart cars are a complex system including perception, decision-making and control. Environmental perception is an important prerequisite for path planning and decision-making control. Camera-based multi-target tracking is one of the key contents of environmental perception. For ADAS such as vehicle driving and pedestrian prediction, etc. System development and path planning for autonomous driving have a major impact. The main task of multiple object tracking (MOT) technology is to simultaneously locate multiple objects of interest in a given video, maintain their identities, and record their trajectories. Through multi-target tracking technology, smart cars can better understand their surrounding environment information and make more pr...

AI Technical Summary

Problems solved by technology

Aiming at the problem that it is difficult for intelligent vehicles to track objects of interest in real time under urban road conditions, the present invention solves the problem that the current algorithm is complex in the post-processing process, and proposes a multi-target tracker for unmanned vehicles based on a self-query mechanism And tracking method, technical scheme of the present invention is as follows:

Images