Efficient Fusion of Dual Lidar Data Streams with Maintained Organized Structure

Abstract This paper presents a method for synchronizing and merging point cloud data from two LiDAR sensors mounted on a vehicle to create a dense, machine-learning-compatible dataset. Two identical Ouster OS1 sensors (32×1024, 20 Hz) were tested in real traffic, and a synchronization algorithm was developed to align frames based on timestamp differences. The merged point clouds were organized into a 64×1024 matrix, with effective handling of missing data to ensure compatibility with neural networks. Computational performance was analyzed across processing steps, highlighting trade-offs between accuracy and speed. Results show that proper synchronization enables efficient sensor fusion, improving perception and robustness in autonomous driving.