Online Multiple Object Tracking Using a Novel Discriminative Module for Autonomous Driving

Abstract

Multi object tracking (MOT) is a key research technology in the environment sensing system of automatic driving, which is very important to driving safety. Online multi object tracking needs to accurately extend the trajectory of multiple objects without using future frame information, so it will face greater challenges. Most of the existing online MOT methods are anchor-based detectors, which have many misdetections and missed detection problems, and have a poor effect on the trajectory extension of adjacent object objects when they are occluded and overlapped. In this paper, we propose a discrimination learning online tracker that can effectively solve the occlusion problem based on an anchor-free detector. This method uses the different weight characteristics of the object when the occlusion occurs and realizes the extension of the competition trajectory through the discrimination module to prevent the ID-switch problem. In the experimental part, we compared the algorithm with other trackers on two public benchmark datasets, MOT16 and MOT17, and proved that our algorithm has achieved state-of-the-art performance, and conducted a qualitative analysis on the convincing autonomous driving dataset KITTI.

Graph Attention Networks and Track Management for Multiple Object Tracking

“…In recent years, MOT has emerged as a research hotspot with wide-ranging applications in fields such as intelligent surveillance, autonomous driving, and behavioral analysis [3]. To associate targets with stability and efficiency, certain studies [4,5] attempted to calculate the distance of the spatial position as a cost matrix by estimating the target's position in the subsequent frame and subsequently applying the Hungarian algorithm for association [6]. OC-SORT [4] predicts the target's state in the next frame using the Kalman filter [7] and calculates the Intersection over Union (IoU) for bipartite graph matching.…”

Automatic Identification of Vehicle Partial Occlusion in Data Collected by Roadside LiDAR Sensors

“…If a target object is occluded, some important information (e.g., shape and color) will be missing and cannot be restored, causing reduced reliability and completeness. If the judgement is not corrected, the accuracy and detection range of object detection, classification, counting, and tracking will all be affected ( 6–16 ). Occlusion can be divided into three categories: full occlusion, partial occlusion, and non-occlusion.…”