Research On Facial Acupoint Recognition AR System Based On Deep Learning

With the outstanding performance of TCM therapies in the fight against the epidemic since 2020,the state has increased its support for the development of TCM.More and more people believe in traditional Chinese medicine,and even start to try scraping,moxibustion and other traditional Chinese medicine methods of home health care.Searching acupoints is an important part of TCM therapy.Whether the acupoint identification is accurate will directly affect the final effect.At present,when people take care of traditional Chinese medicine at home by themselves,it is basically amateur operation,which inevitably leads to deviations in the positioning of acupoints.Therefore,this paper studies the accuracy and real-time performance of acupoint recognition.Firstly,a facial acupoint recognition algorithm based on deep learning is proposed.Secondly,based on this algorithm,a real-time AR system of moving end point positioning is designed and implemented.The following is the main content of this paper:（1）MTCNN face detection algorithm is improved from two aspects.Firstly,the network structure of MTCNN is optimized.The calculation part of face key point detection in P-NET,R-NET and O-NET is removed.Thereby reducing the complexity of the algorithm.Secondly,improve the size of the minimum face image in the image pyramid.Reduce the number of layers in the image pyramid.Thereby reducing the amount of calculation of face detection.Comparison experiments show that the improved MTCNN is faster than the original MTCNN in face detection tasks.（2）The PFLD key point detection model was improved.The lightweight convolutional neural network Mobile Net V3 is proposed as the backbone network of the PFLD model.This can increase the accuracy and efficiency of detection.At the same time,the data set of face key point detection was amplified by rotation and mirror operation to further improve the robustness of PFLD model.Through the comparison test between the improved PFLD model and the original PFLD model.It was found that the improved PFLD model detection speed and accuracy were improved.（3）This paper studied the position relationship between acupoints and the coordinates of facial key points and proposed the calculation method of the coordinates of facial acupoints combined with the bone degree measurement method of traditional Chinese medicine,and finally realized the accurate positioning of more than 20 facial acupoints.（4）Based on the above algorithm research,a multi-functional facial acupoint realtime recognition AR system running on smart phones is developed.The facial acupoint recognition AR system researched and designed in this paper has the characteristics of high accuracy,strong popularity and simple operation.It has important practical significance for common people’s identification of acupoints and inheritance of traditional Chinese medicine culture.