A new approach of obstacle fusion detection for unmanned surface vehicle using Dempster-Shafer evidence theory

Abstract

Aiming at the demand of obstacle collision avoidance for unmanned surface vehicle (USV) at near-field, a new approach of obstacle fusion detection for USV using Demperster-Shafer (D-S) evidence theory was established. The grid map was constructed as the environmental model of USV. The multi-sensor data was used as evidence to determine the attribute of each grid. According to the detection accuracy of each sensor in different perception area, the mass functions of evidence were assigned for each grid. The d-S combination rule was used to make a final fusion decision of the grid attribute. In this way, the obstacle fusion representation for USV was accomplished. Furthermore, a validation experiment was carried out using the developed multi-sensor obstacle detection system. In the experiment, the perception data of three-dimensional lidar, millimeter wave radar, and stereo vision were acquired from four typical obstacle scenes at sea. The qualitative and quantitative comparison results between multi-sensor fusion and single sensor detection show that the obstacle fusion detection method can make use of the complementarity between different sensors to enrich the detection information of obstacle, and it can effectively avoid the single sensor false detection for enhanced situational awareness of USV. The novel fusion detection method shows greater advantage in the reliability of obstacle detection.