Abstract
LiDAR (Light Detection and Ranging) sensors are widely used in self-driving cars with awareness of the surrounding environment. However, the LiDAR sensor is sensitive to harsh weather conditions that cause the collected data to be distorted. These types of weather reduce the safety of self-driving cars. The harsh weather conditions also cause missing points problems on the point clouds, and it causes the performance of 3D object detection to reduce. Therefore, we propose a new method using probability estimation, which includes a Deep Mixture of Factor Analyzers (DMFA) and a Miss-Convolution layer, to recover missing points caused by snow. The proposed work outperforms models which perform well in normal conditions. In summary, snow often causes detection errors for 3D modern detectors. By recovering missing points in the point cloud, we significantly make the performance of the 3D detector better in snowy weather conditions.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have