Acupuncture therapy is one of the main form of treatment in Traditional Chinese Medicine (TCM). Based on different patient symptoms, needling or massage is applied to the corresponding acupuncture points to relieve the symptoms. However, given the large number of acupoints and the complexity of their specificity, it is difficult for one to remember the location and function of each acupoint without professional training. Due to the limitation of physiological structure, it is difficult for people to massage the acupoints on the back by themselves. In this work, we propose a system that can estimate acupoint location by back image, and retrieve the 3D coordinate of the acupoint in real world using the depth camera, and transform the coordinate into robot arm space so that the robot arm can automatically localize the acupoint.

In our work, through learning the acupoint localization method summarized by TCM, we propose an approach for back acupoint estimation by leveraging a mean back model consisting of landmark points and spine locations. We build a spine—acupoint relation model that records the relative distance between acupoint and its reference vertebra. Also, uArm Swift Pro is incorporated into the system to automatically localize acupoints on the back.