ITSC 2025 Paper Abstract

Paper WE-LA-T7.1

Zimmer, Walter (Technical University of Munich (TUM)), Pranamulia, Ramandika (Technical University of Munich, Germany), Xingcheng, Zhou (Technical University of Munich), Zhang, Jiajie (Technical University of Munich), Wei, Chuheng (University of California, Riverside), Greer, Ross (University of California, San Diego), Berrio Perez, Julie Stephany (University of Sydney), Liu, Mingyu (Technical University of Munich), Trivedi, Mohan M. (University of California at San Diego), Knoll, Alois (Technische Universität München), Caesar, Holger (TU Delft)

PointCompress3D: A Point Cloud Compression Framework for Roadside LiDARs in Intelligent Transportation Systems

Scheduled for presentation during the Regular Session "S07c-Smart Infrastructure and Data-Driven Sensing for Intelligent Mobility" (WE-LA-T7), Wednesday, November 19, 2025, 16:00−16:20, Coolangata 1

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 19, 2025

Keywords Cloud and Edge Computing Integration in ITS for Real-time Traffic Data Processing, IoT-based Traffic Sensors and Real-time Data Processing Systems, Lidar-based Mapping and Environmental Perception for ITS Applications

Abstract

In the context of Intelligent Transportation Systems (ITS), efficient data compression is crucial for managing large-scale point cloud data from roadside LiDAR sensors. There is strong demand for compact storage, streaming, and real-time object detection due to the high data volume and need for rapid processing in time-critical applications. This work introduces PointCompress3D, a novel compression framework tailored for roadside LiDARs. It addresses the challenges of compressing high-resolution point clouds in real time while preserving accuracy and sensor compatibility. We adapt and integrate three state-of-the-art compression methods and evaluate them on real-world TUM Traffic datasets. Moreover, we deploy our framework on a live ITS test bed and test it under real traffic conditions. After fine-tuning, we achieve 10 FPS with compression sizes under 105 KB, a reduction of 50 times, while maintaining object detection performance comparable to the original data. We achieve a point-to-plane (d2) PSNR of 94.46 in extensive experiments and ablations on the TUM Traffic datasets. Code and video results are available on our website: https://pointcompress3d.github.io.