NBA3D: Neighbor-Based Confidence Adjustment for 3D Rare Object Detection Using LiDAR

As humans cause most severe car crashes [71] , lawmakers have been enforcing mandatory driver assistance systems for vehicles such as electronic stability control. Advanced driver assistance requires reliable environmental perception through cameras, radar, and LiDAR sensors. When installed on road infrastructure, in addition to vehicles, these sensors facilitate autonomous transport because they allow foresight of traffic events from a more convenient perspective. Provided that vehicles receive complete environment information externally from road infrastructure via wireless communication, car manufacturers could save on some costly on-board sensors. To obtain complete environment information, the sensor data must be automatically analyzed by intelligent object detection software. This research work produces a modular unsupervised approach that does not require training on sample data and processes LiDAR data at near 50 Hz while extracting objects and their 3D information. The presented solution is a suitable code base, which can be further improved by related research works, examined in a comprehensive literature review.