Abstract—Autonomous vehicles, known as self-driving cars, rely heavily on the 3D LiDAR sensor alongside other sensors to navigate autonomously. 3D LiDAR sensor produces sparse point clouds which makes classifying and detecting objects inside them difficult. However, the introduction of deep learning made it possible which led to various types of 3D object detectors for 3D pointclouds. Existing 3D object detectors can identify cars, pedestrians, and cyclists. However, on the road with the abundant presence of heavy vehicles like trucks and vans, this paper explores the potential to retrain an existing 3D object detector, PointPillars, using transfer learning, to also detect vans and trucks in the 3D LiDAR. By training it to do so, autonomous vehicles can recognize and respond appropriately to heavy vehicles during navigation. Then by using Average Precision metric, the results showed an average performance for "Van" but a below-average performance for "Truck". Interestingly, there's an improvement on the Average Precision score for the "Car" and "Cyclist" classes with a drop on the "Pedestrian" class. The modified PointPillars achieved mean Average Precision (mAP) scores of 35.53 and 34.57 for the “Van” class using Bird's Eye View (BEV) and 3D metrics, respectively, at IoU = 0.5 and at "Moderate" difficulty. On the other hand, the mAP scores for the "Truck" class are 10.09 and 4.46 using BEV and 3D metrics, respectively at IoU = 0.5 and at "Moderate" difficulty.

Keywords—3D machine learning, LiDAR, transfer learning, Autonomous Vehicles (AV)

Cite: Ericsson Yong, Muhammad Aizzat Zakaria, Mohamad Heerwan Peeie, and M. Izhar Ishak, "Van and Truck Classifications of 3D LiDAR Perception for Autonomous Vehicle Navigation," International Journal of Mechanical Engineering and Robotics Research, Vol. 14, No. 3, pp. 276-281, 2025. doi: 10.18178/ijmerr.14.3.276-281

Copyright © 2025 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).