Efficient Traversability Mapping Based on Single Camera and 3D LiDAR

Mobile robots that are deployed both indoors and outdoors require the capability of recognizing their environment in real-time to improve their autonomous navigation. Many researchers study the leverage of cameras and light detection and ranging (LiDAR) sensors in combination to generate a representation of the environment. LiDAR is widely adopted for its ability to create high-precision maps in mobile robots and autonomous vehicles. With advanced deep learning techniques, various camera perception methods such as semantic segmentation, object detection, and classification represented remarkable performance. In this paper, we propose an efficient traversability map that fuses the recognition capability of cameras with the accurate mapping of 3D (LiDAR) sensors. Finally, we show an approximate F1 score performance of 0.83 compared to the LiDAR-based map generation method.