Abstract

Info

Title:: Obstacle avoidance in simulated prosthetic vision based on SOLOv2-RS

Author(s):: E Ning1; WANG Jing1; 2; ZHOU Xianglong1; ZHAO Rongfeng1; HE Haiyang3; 1. College of Information Technology, Shanghai Ocean University, Shanghai 201306, China 2. Key Laboratory of Fisheries Information, Ministry of Agriculture and Rural Affairs, Shanghai 201306, China 3. Shanghai SenseTime Intelligent Technology Co., Ltd., Shanghai 200231, China

Keywords:: Keywords: visual prosthesis obstacle avoidance SOLOv2-RS ResNeSt hierarchical optimization processing

Abstract:: Abstract: Aiming at the obstacle avoidance in simulated prosthetic vision, an improved instance segmentation model SOLOv2-RS is proposed for providing a basis for implant recipients to accurately perceive the relevant instance objects of navigation tasks in low-resolution prosthetic vision. According to the visual attention mechanism, the distance from the center of the visual field and the target scale are adopted as the importance calculation criteria for each instance, and the obtained importance score is used as the basis for the hierarchical representation of the obstacles to be avoided. Meanwhile, edge information is used to cue the tactile paving, and it is morphologically inflated for avoiding the edge information loss caused by the limited phosphene. The prosthetic vision simulation results demonstrate that the hierarchical optimization processing strategy for simulated prosthetic vision can effectively achieve the optimal representation of tactile paving and obstacles, thus facilitating the implant recipients to accomplish outdoor obstacle avoidance tasks more efficiently, and providing ideas for the research on the image processing of visual prosthetic devices.