LiDAR-Based Multi-Object Pedestrian Tracking Using an Extended Kalman Filter for Autonomous Driving Otonom Sürüş için Genişletilmiş Kalman Filtresi Kullanarak LiDAR Tabanli Çoklu Nesne Yaya Takibi

Sahin I. S.

34th Signal Processing and Communications Applications Conference, SIU 2026, İstanbul, Turkey, 7 - 10 July 2026, (Full Text)

  • Publication Type: Conference Paper / Full Text
  • Doi Number: 10.1109/siu71813.2026.11636704
  • City: İstanbul
  • Country: Turkey
  • Keywords: Autonomous Vehicles, Extended Kalman Filter, LiDAR, Multi-Object Tracking, Pedestrian Tracking, ROS2
  • Yıldız Technical University Affiliated: Yes

Abstract

For autonomous driving systems, reliable pedestrian tracking is a critical requirement, especially in safety-critical urban environments where sensor measurements are noisy and intermittent. While object detection methods provide location information, they do not provide temporal consistency and velocity information. This study aims to implement LiDAR-based multi-object pedestrian tracking using an Extended Kalman Filter (EKF). Pedestrian candidates are primarily detected through the clustering of 3D LiDAR point clouds, and the resulting cluster centers are used as location measurements. Each detected pedestrian is tracked by an independent EKF based on a constant velocity motion model, providing robust state estimation and temporal consistency under measurement uncertainty. A data association mechanism based on Mahalanobis distance is used to match incoming measurements with existing tracks, while an adaptive track management strategy is implemented to prevent false obstacle persistence when the number of detected pedestrians decreases. The proposed system has been implemented in ROS2 and visualized in real-time using RViz. Tracking performance was evaluated using the root mean square error (RMSE) metric, and it was shown that EKF significantly reduced measurement noise and improved tracking stability compared to raw cluster observations. The results confirm that EKF-based multi-object tracking provides an effective and computationally efficient solution for pedestrian tracking in autonomous vehicle applications.