LIDAR Fault Detection Using Synthetic Point Cloud Training

The patent creates synthetic LIDAR point cloud data that copies the characteristics of real LIDAR data through simulation. Normal point cloud data is generated from 3D model data of vehicles and surrounding objects, and abnormal data is created by introducing various failure modes into this synthetic data. This allows the detection model to be trained on diverse virtual scenarios without requiring physical test patterns or targets in specific environments.

The patent varies multiple parameters in the synthetic data generation process including object positions, distances, orientations, weather conditions, and LIDAR failure modes to create diverse training datasets. By changing these parameters systematically, the model learns to detect malfunctions across various environmental conditions rather than being limited to specific test scenarios.

Instead of collecting real abnormal LIDAR data through actual failures or test patterns, the patent copies the essential characteristics of abnormal data by introducing simulated failure modes into synthetic point cloud data. This includes adding noise, occlusions, and other artifacts that mimic real malfunction conditions without requiring actual LIDAR failures or extensive field testing.

The patent uses computationally inexpensive synthetic data generation methods rather than expensive and time-consuming real-world data collection. The synthetic datasets can be generated quickly through simulation and can be discarded or regenerated as needed, unlike real abnormal data which requires extensive field deployment and actual failure occurrences.