ITSC 2024 Paper Abstract

Paper ThBT4.6

Zeng, Zifan (Technical University of Munich), WANG, PENG (Huawei Technologies Co., Ltd., RAMS Lab), zhang, qunli (Huawei Technologies Duesseldorf GmbH), Liu, Shiming (Huawei Technologies Co., Ltd.), Liu, Feng (Huawei Technologies Duesseldorf GmbH), Feng, Bin (shanghai huawei technologies co. ltd)

Synthetic Point Clouds Generation for ToF LiDAR under Rainy and Foggy Weathers Using Monte Carlo Based Signal-Environment Interaction Simulation

Scheduled for presentation during the Regular Session "Synthetic datasets in perception" (ThBT4), Thursday, September 26, 2024, 16:10−16:30, Salon 7

2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), September 24- 27, 2024, Edmonton, Canada

This information is tentative and subject to change. Compiled on October 3, 2024

Keywords Data Mining and Data Analysis, Sensing, Vision, and Perception, Driver Assistance Systems

Abstract

The safety of autonomous driving systems (ADS) is of paramount importance and heavily relies on precise sensors and robust perception algorithms. However, these are often prone to adverse weather conditions such as rain, fog and snow, which significantly affect the sensor signals and, consequently, the perception module’s understanding of the environment. Compared to clear weather data, collecting driving data in harsh weather such as rainy and foggy days is costly and challenging. This paper proposes a novel method for synthetic Time-of-Flight (ToF) Light Detection and Ranging (LiDAR) point cloud generation under rainy and foggy weathers using a Monte Carlobased signal-environment interaction simulation. By establishing a physical model of laser interaction with particles in adverse environment and conducting multiple random simulations, synthetic point clouds are generated from the existing clear weather dataset accordingly. Experimental results demonstrate promising effectiveness and application prospects in the field point cloud data augmentation in uncommon weather, which fosters the tailored capability to improve the robustness of perception algorithms through training thereby enhancing the safety of autonomous driving systems.