Acupuncture Robot Detects Acupoints Based On Improved Graph Convolution

Acupuncture,as an important part of traditional Chinese medicine,has played an important role in the fight against COVID-19.However,as the number of patients in need of acupuncture therapy has increased,the number of acupuncturists has become insufficient.Therefore,the research of acupuncture robots is put on the agenda.In this paper,two stages of acupuncture and moxibustion robot,three-dimensional reconstruction and localization of acupoints,are studied.In the stage of 3D reconstruction,this paper studies two problems existing in the current research.Firstly,the current research on 3D reconstruction of monocular images mainly focuses on human posture estimation.However,there are relatively few studies on the problems of body reconstruction.Secondly,there are relatively few methods to estimate human body shape parameters.In the stage of acupoint localization,this paper mainly studies three aspects.First of all,there are relatively few researches in the industry on acupuncture robots completing the task of locating acupoints on 3D human models,and few relevant data sets can be found at present.For some acupoints with very unclear local features(such as the Dazhui acupoint targeted in this paper),there is no relevant data set to support the deep learning method to solve this problem.Secondly,the experiment shows that it is difficult to accurately locate the acupoint by using the existing method directly.Finally,experiments show that when various deep learning methods used in this paper complete the task of acupoint positioning,the prediction points are always difficult to fit well on the surface of the model.In order to solve the problems in 3D reconstruction of monocular images,this paper firstly proposed a data set generation method based on projection,which can obtain the attitude parameters of projection better and provide more accurate data for training.Secondly,this paper proposes a human body shape parameter estimation method based on generation assistance.Inspired by the multi-angle human body three-dimensional work,this method generates the output that can assist body shape parameter estimation for the whole model by learning the features of the input image.Experiments show that this method can improve the ability of estimating body shape parameters for different existing methods.For the problem of acupoint positioning in the three-dimensional human point cloud model,this paper firstly systematically combs various methods of acupoint positioning completed by the acupuncture robot.Secondly,a dataset containing about 10,000 human models with different body parameters marked with Dazhui point location was constructed.Thirdly,this paper proposes a point cloud deep learning model based on graph aggregate projection and its corresponding adjacency matrix generation method.Finally,an optimization method based on the center of gravity method is proposed.Experiments show that the point cloud deep learning model based on graph aggregation projection can effectively predict the location of Dazhui point,and the graph aggregation projection in the proposed method can effectively assist in estimating the location of Dazhui point.In addition,the optimization method based on the barycentric method can also effectively fit the predicted points to the human body surface.