Human Motion And Posture Behavior Identification Under Simulated

Eighty percent of the information humans perceived from the outside world comes from visual perception.The visual perception caused by retinal pigment degeneration and age-related macular degeneration could cause people to lose this way of obtaining information.In recent years,the restoration of the visual function for blind patients has mostly focused on the research of visual prostheses.Visual prosthesis is electrically stimulating the patient's remaining available visual nervous system to make patients have photophantoms to achieve visual feeling.Human motion recognition is the most important part of human daily behavior,so it is a very important part of human behavior recognition in visual restoration.Due to the limitations of current technology,the number of microelectrodes implanted into the human body is limited.The number of electrodes for visual prosthesis implanters to achieve maximum visual experience is a hot spot in visual prosthesis research.By extracting significant human behavior information in a suitable way,it can provide patients with more visual information with a limited number of electrodes.In this study,human skeletal points and their connections were extracted and pixelated,aiming to greatly improve the human behavior recognition rate at low resolution.Using(Partial Affinity Threshold)Part Affinity Fields' human pose estimation method to obtain the position of bone key points and their connection information,the bone points and their connections without background were extracted to obtain the skeleton image,and processed into 16×16,24×24,32×32,48×48 pixelated video images at statistical resolution,statistical analysis of the recognition accuracy of 40 subjects at different resolutions were recored and analysed.The experimental statistical results showed that,under the vision of the artificial prosthesis,as the resolution increases,the accuracy of human behavior recognition increased;the accuracy of the human behavior recognition has no significant difference in gender and visual experience with or without the artificial prosthesis.The experimental results were compared with the graphs processed by the FT(Frequency Adjusted Significant Region Detection)algorithm and the SR(Residual Normal)saliency map algorithm.The FT algorithm used the center-periphery operator of the color feature to obtain a saliency map,which has a small amount of calculation and is also effective in extracting the significance of the target boundary.The SR model can obtain the residual spectrum of the input image in the spatial domain by analyzing the log spectrum of the input image,and then use the fast method to construct the relevant saliency map in the spatial domain.The test results of the model on natural and artistic images showed that the method has the characteristics of high computational efficiency and good robustness.The FT algorithm and the residual general model were used for saliency map extraction and pixelation processing,and the processed pixelated image was compared with the skeletonized image after pixelation processing.The pixelated image after skeleton extraction has a low resolution recognition accuracy rate.Has significant advantages.