Development and Evaluation of a LiDAR- and Camera-Based Multi-Object Tracking Framework for Robust Traffic Flow Measurement under Adverse Weather Conditions

Multi-object tracking (MOT) facilitates mobile robots in executing informed motion planning and navigation by accurately localizing objects within a three-dimensional space. Originally designed for automotive applications, MOT systems have found extensive application in smart city initiatives aimed at monitoring traffic flow. Most existing methods rely on either LiDAR sensors for 3D tracking or cameras for tracking objects in 2D space. However, each has its limitations, making them less effective in certain scenarios.
This thesis aims to develop the Multi-Object Tracking(MOT) frameworks that integrate the data from both camera and LiDAR sensors to ensure reliable trajectory measurement of multiple traffic participants across diverse weather conditions. The data used in this work was acquired from a mobile Roadside Unit (RSU) developed in a previous project, which is equipped with two high-resolution LiDAR sensors, an RGB camera, and high-fidelity weather sensors. The object detection algorithm for the LiDAR sensor was provided by an industrial partner, while the popular YOLOv8 algorithm was used for camera-based detection, which serves as primary inputs for the object tracking framework.