LIDAR and monocular based overhanging obstacle detection

Abstract

This paper presents an improved method for the detection of obstacles in the trajectory of autonomous ground vehicle (AGV). The novel approach requires fewer calculations, i.e. less computational time. Obstacle detection algorithms were investigated, in order to perform safe motion control, in an environment with unknown overhanging obstacles. We describe a two dimensional (2D) laser sensor application, and optimal sensor configurations for mounting a monocular camera to monitor path ahead clearance. Two different sensors are used, a vision sensor and a scanning laser, Light Detection and Ranging (LIDAR). While LIDAR measures the precise distance to the object, it cannot detect low objects and overhanging obstacles due to its predefined, constant, scanning height and angle. In contrast, vision sensor provides 2D scenery information with relatively poor distance information. To compensate for the drawbacks of these two sensors, the sensor fusion method for obstacle detection of AGV is proposed. Size expansion cue algorithm is deployed to achieve that goal. Proposed method is validated experimentally.