Motion Recognition Based On Video Saliency Under Simulation Prothesis Vision

In medicine,retinitis pigmentosa(RP)and Age-related macular degeneration(AMD),cause irreversible blindness and are still incurable.The visual prosthesis gives patients hope to see the light again.With the development of science and technology,many research groups in the world have implanted visual prosthesis in different areas,and made remarkable achievements in clinical application.However,the resolution of visual prosthesis is still far lower than that of natural vision.Therefore,this research focus on to find the optimal image processing strategy under visual prosthesis.In this paper,two psychophysical experiments were designed to simulate the visual recognition of prosthesis.In the first experiment,three image processing strategies,including two traditional edge extraction algorithms and a saliency aware region detection algorithm(SAG)based on perceptual detection were used to process 30 motion videos in ucf-101 database,then match the processed videos with different resolutions templates(48 × 48,64 × 64,128 × 128).The time and recognition accuracy of the subjects in different resolution and different image processing strategies were recorded and analysised.The experimental results showed that compared with the other two traditional edge extraction algorithms,SAG improved the accuracy in the three image processing strategies significantly.This is because SAG algorithm can provide more gray information to the subjects,and filter the redundant information in most pictures automatically.In the case of low resolution,more gray-scale information can better help the prosthesis wearer identify the action information.In the second experiment,three image processing strategies,including a direct pixel method,a traditional edge extraction algorithm and a face to face Chinese character conversion system based on FaceNet,were used to process 36 videos(18 standing groups and 18 sitting groups)from four students of Inner Mongolia University of science and technology and match the processed videos with different simulation resolutions(36 × 36,48 × 48,64 × 64,).The time and accuracy of recognition under different resolution and different image processing strategies Record and analysis.The experimental results showed,among the three image processing strategies,the proposed image processing strategy can help the subjects to obtain higher recognition accuracy in a shorter time.When the resolution was 36 × 36,the recognition accuracy using our strategy has reached 100%.Therefore,in the second experiment,the recognition experiment using our strategy when the resolution was reduced to 24 × 24 were also conducted.The experimental results showed that the average recognition accuracy was 61.10%.Finding the best image processing strategy of video motion recognition and character assisted recognition under the condition of low resolution could help the prosthesis wearer adapt to real life better and complete independent activities as soon as possible.Although the simulation visual prosthesis could provide researchers with a way to simulate the vision of the prosthesis wearer for experiments,compared with the irregular optical illusion and resolution loss in the actual visual perception of the wearer,the visual perception provided by the simulation visual prosthesis is still in idealization.These factors will be considered and discussed in the future researches.