Multi-Object Tracking and Speed Bump Detection based on LiDAR Point Clouds =

Multi-object tracking and road feature detection are critical perception tasks in Advanced Driver Assistance Systems (ADAS), as they directly affect perception robustness and driving safety. This paper proposes algorithms for target tracking and speed bump detection based on solid-state LiDAR point clouds. For multi-object tracking, an Extended Kalman Filter (EKF) with a Constant Turn Rate and Velocity (CTRV) motion model is used to estimate full target states including position, velocity, heading, and yaw rate. By performing inter-frame data association and state estimation on solid-state LiDAR data, the method enables robust tracking of dynamic targets. A Markov chain–based track management strategy ensures reliable initiation, maintenance, and termination of tracks in complex traffic environments. For speed bump detection, this paper proposes a method based on local point cloud height variations. Edge points are first extracted from ground point clouds using local height differences and then clustered to derive geometric features including position, size, slope, and confidence. These features are incorporated into a geometric decision model designed for sparse LiDAR data, enabling accurate and reliable speed bump localization. Validated detections are published via ROS topics, integrated into the Autoware perception pipeline, and transmitted through CAN messages for vehicle control support. The algorithms were developed in ROS2 Humble, deployed on NVIDIA® Jetson™ AGX Orin platform, and tested on a PIX-KIT 2.0 platform with verification in the Autoware environment. Real-world experiments demonstrate accurate tracking, robust speed bump detection, and efficient runtime performance. This work provides a practical perception solution for low-cost intelligent driving systems and broadens the application of LiDAR in static road structure recognition.