Research On Blind-assisted Visual Algorthm Based On FPGA

People often say "eyes are the windows of the mind". Because of congenital or acquired factors, thousands of people can't see anything. The daily life of the blind have many restrictions, such as walking can't aware of the danger in time, which makes lives difficult for the blind. With the rapid development of computer technology and communication technology,the auxiliary equipment for the blind has developed fast, also more and more researchers focus on related technologies.This paper mainly studies two aspects of detection: the dynamic target detection and the static target edge detection, and completes the following work: Firstly, this paper describes the principle of the traditional algorithms and analyzes which scene is best for the algorithms, and adopts the improved algorithm for the blind walking auxiliary. Using the ViBe algorithm combined manual way which needs to count pixels number to update and reinitialize the background model in dynamic target detection, using adaptive threshold method in the static target edge detection. Secondly I use MATLAB to compare the improved algoritlhm with the traditional algorithms, and use two ways which includes objective data evaluation and the subjective observation to evaluate them respectively. The Recall of improved dynamic target detection algorithm is 10% higher than the original one, and the Precision is 6% higher than the original one. The static edge extraction algorithm can effectively extract the required edge information. Thirdly I design circuit for the improved detection algorithms with Verilog and simulate it with ModelSim. Finally I build target detection system, from the camera configuration to VGA display, to verify the overall system on the development board.Comparing hardware implementation results with MATLAB processing results, the hardware processing results are compared with the software processing results at the pixel maximum relative error of 4.3%, the hardware implementation to achieve the design requirements.The result of target detection experiment shows that the dynamic target detection which adopts ViBe algorithm combined manual way that needs to count pixels number to update and reinitialize the background model, that makes the target detection more suitable for actions shot.Static target edge detection with adaptive threshold can reduce interference of irrelevant information. The hardware implementation of the target detection meets the real-time requirement, and enrich the research contents of image processing in the visual prosthesis.