Research On Acupoint Recognition And Localization Technology Based On Machine Vision

In recent years,the development of intelligent rehabilitation diagnosis and treatment equipment based on acupoints has received increasing attention.The recognition and localization of acupoints highly rely on the experience of doctors,becoming a bottleneck in the development and promotion of intelligent rehabilitation equipment based on acupoints.To address the above issues,this thesis conducts research on acupoint recognition and localization technology based on machine vision,laying the foundation for the subsequent development of intelligent rehabilitation equipment.(1)Based on the summary and analysis of traditional Chinese medicine theoretical knowledge,combined with the characteristics of modern machine vision and neural network technology,human acupoints are divided into two categories: obvious visual features and blurred visual features.Based on their respective characteristics,feature point assisted localization methods and region segmentation localization methods are proposed respectively;Establish an experimental platform using Kinect depth camera,rokae 7 robotic arm,and treatment bed equipment,while completing the calibration of the depth camera,registration of depth information with pixel coordinates,and mapping of pixel coordinate system to robotic arm coordinate system.(2)For acupoints with obvious visual features,this thesis proposes a feature point assisted localization method to achieve recognition and localization of such acupoints.Divide the human body into multiple subgraphs and use Res Net50＿our classification network model to classify,find the subgraph where the feature points are located,and then use the Canny contour detection algorithm and CPDA＿our corner detection algorithm achieves automatic recognition and localization of feature points,and finally combines the position relationship between acupoints and feature points in traditional Chinese medicine finger-length measurement to obtain the pixel coordinates of such acupoints.(3)For acupoints with blurred visual features,this thesis proposes a region segmentation localization method to achieve recognition and localization of such acupoints.Select Dazhui acupoint,Tianzhu acupoint,Zhushu acupoint,Jianzhen acupoint,Guanyuanshu acupoint,and spinal spinous process as key acupoints and create a semantic segmentation dataset based on them.Then,use the SK-UNet semantic segmentation network model to segment the human body,and obtain the pixel coordinates of these acupoints based on the relationship between the segmentation results and the position of the key acupoints.(4)Acupoint recognition and localization verification experiment.This thesis recognizes and locates 37 common acupoints,considering the lack of unified evaluation indicators for acupoint localization,referring to the requirements of acupuncture and moxibustion,moxibustion,massage and other treatment methods on the accuracy of acupoint localization under different error radii,the error radii of 5mm,10 mm,15mm and20 mm were taken as the evaluation criteria for the accuracy of acupoint localization.The average error of acupoint pixel coordinate localization is4.68 mm,and the localization accuracy is 59.80%,91.95%,96.35% and98.83% when the error radius is 5mm,10 mm,15mm,and 20 mm,respectively;The average error of localization through the robotic arm is4.89 mm,and the localization accuracy is 55.10%,87.16%,92.06% and96.69% when the error radius is 5mm,10 mm,15mm and 20 mm,respectively.The experimental results indicate that this scheme can effectively complete the task of acupoint recognition and localization,meeting the needs of subsequent intelligent rehabilitation treatment.