Gait Recognition In Simulated Prosthesis Vision

Most of human’s external information comes from the eyes,and obtaining image information through vision is an important function of human learning and life.China is the country with the most blind people in the world.The main cause of blindness is retinal diseases,and the most common retinal diseases are incurable Retinitis Pigmentosa and primary Age-related Macular Degeneration.Retinal prosthesis is considered to be one of the feasible methods to restore functional vision at this stage.It restores partial visual function by implanting electrode arrays to induce phosphenes in patients who are blind due to retinal diseases.Based on this,various simulation prosthesis experiments aim to optimize the visual presentation of the implanter.At present,the research on identity information recognition under simulated prosthesis vision mostly focuses on face recognition at short distances,while at long distances,the biometric feature of human motion gait can be used for identity recognition.In order to explore the recognition of identity information through gait motion information under simulated prosthetic vision,a psychophysical experiment of human gait movement information recognition under simulated prosthetic vision was designed.The pedestrian videos in the experiment come from 8 males and 8 females in the CASIA-B gait database.Each pedestrian has three motion angles of 0°,54° and 90° and two walking states: normal state and overcoat state.Firstly,use algorithms such as object detection to extract and optimize the foreground of pedestrian motion in the normal state,and process it into pixelated videos at three resolutions(32×24,48×36,64×48).The recognition accuracy and correct identification time of gender classification and identification from three perspectives of 20 subjects(10 males and 10 females)under three resolutions were recorded.The experimental results show that it is feasible to identify by gait information under simulated prosthesis vision,and the recognition accuracy increases with the improvement of resolution,and normal recognition can be performed at 48×36 resolution.The recognition speed and accuracy are the best under the 90 ° angle of view.At the same time,there is no significant difference between the gender of subjects and pedestrian.Secondly,the same method was used to process the video in the state of wearing a thick coat into pixelated video at two resolutions of 48×36 and 64×48,and recorded and statistically analyzed the gender classification of 20 subjects under different resolutions and different viewing angles.and identification accuracy.The experimental results show that under the two resolutions,the accuracy of gender classification does not change much,and the accuracy of identification is slightly improved;among the three viewing angles,the difference in recognition accuracy is not significant;comparing the two states,in the case of wearing a jacket,the accuracy of gender classification decreases,especially The gender of women is more difficult to identify,and the accuracy of identification has dropped significantly.The results of this study can provide strong data support for clinical trials of visual prosthesis implantation.