Synthetic LiDAR Point Cloud Data Generation Tool and Deep Learning Validation
2022-01-0177
03/29/2022
- Event
- Content
- LiDAR sensors are set to become common in automated driving due to their high accuracy. However, LiDAR processing algorithm development suffers from lack of training data, partly due to the sensors' high cost. The few public datasets (e.g. KITTI) offer poor coverage of edge case scenarios, whereas these are essential to address to support safe self-driving. We address the unmet need for abundant, high-quality LiDAR data with the development of an artificial LiDAR point cloud generation tool and validate this tool's performance against the PIXOR model, which was trained on the KITTI dataset. The tool makes use of a single camera raycasting process to generate synthetic data and is used in conjunction with filtering techniques to create segmented and filtered object class-specific datasets . The tool has the potential to support the low-cost generation of accurate bulk data in support of training advanced self-driving algorithms, with configurability to simulate commercial and yet-to-be-developed LiDAR with varying channels, range, vertical & horizontal fields of view and angular resolution. The simulator is developed using the Unity Game Engine in conjunction with free and open-source assets and will be shared with the AV community as an open source project.
- Citation
- Karur, K., Pappas, G., Siegel, J., and Zhang, M., "Synthetic LiDAR Point Cloud Data Generation Tool and Deep Learning Validation," SAE Technical Paper 2022-01-0177, 2022, .