Methods for Improving Point Cloud Authenticity in LiDAR Simulation for Autonomous Driving: A Review

Abstract

Collecting LiDAR data for autonomous driving using real vehicles is costly, scenario-limited, and challenging to annotate. Simulated LiDAR point clouds offer flexible configurations, reduced costs, and readily available labels but often lack the realism of real-world data. This study provides a comprehensive review of methods to enhance the authenticity of simulated LiDAR data, focusing on simulation scenarios, environmental conditions, and point cloud features. Additionally, we discuss verification techniques, including direct and indirect methods, to assess authenticity improvements. Experimental results demonstrate the effectiveness of these techniques in enhancing perception algorithm performance. The paper identifies challenges in simulating LiDAR data, such as accuracy discrepancies, brand adaptability, and the need for comprehensive evaluation metrics. It also proposes future directions to bridge the gap between simulated and real-world data, aiming to optimize hybrid training models for improved autonomous driving applications.