Recognition of virtual maze scene under simulated prosthetic vision

To help blind people obtain the optimal vision in low resolution effectively, real-time maze pathfinding experiment under simulated prosthetic vision was carried out to find the optimal real-time scene processing strategy using the limited number of electrodes. Simple (5×5), medium (8×8) and complex (11×11) maze scenes were built by Unity. Then, binarization, color inversion and matching to 32×32 resolution phosphene template were performed as image processing strategy to get useful scene information from the camera and realize real-time pixelization. In the experiment, subjects were asked to complete 45 and 60 degrees viewpoint of the maze pathfinding task at 32×32 resolution. The time of finding entrance, maze pathfinding and the accuracy rate of completing maze from different view angles under the same resolution were recorded and analyzed to determine the optimal viewing angle. As the results, Subjects were able to accomplish the most complicated maze routing task at 32×32 resolution, the time of finding entrance under 45 degrees viewpoint and 60 degrees viewpoint were 14.38 seconds and 19.30 seconds, the time of 45 degrees viewpoint was significantly less than that of 60 degrees. The time of the maze pathfinding under two view angles were 125.39 seconds and 170.14 seconds, and the time of 45 degrees viewpoint was less than that of 60 degrees. Therefore, the 45 degrees viewpoint was significantly better than the 60 degrees visual angle and provided the best viewpoint for visual prosthesis to complete maze wayfinding tasks.