Multi-Modal Image Segmentation for Obstacle Detection and Masking

Lee, Cheng-Lung, Hongyi Zhang, Hong Nguyen, Yu-Ting Wu, Christopher Smalley, Mohammad Utayba, and Mark Paulik

A novel multi-modal scene segmentation algorithm for obstacle identification and masking is presented in this work. A co- registered data set is generated from monocular camera and light detection and ranging (LIDAR) sensors. This calibrated data enables 3D scene information to be mapped to time- synchronized 2D camera images, where discontinuities in the ranging data indicate the increased likelihood of obstacle edges. Applications include Advanced Driver Assistance Systems (ADAS) which address lane-departure, pedestrian protection and collision avoidance and require both high-quality image segmentation and computational efficiency. Simulated and experimental results that demonstrate system performance are presented.