Research On Head Pose Recognition In Rehabilitation Training Scene

Cervical spondylosis has become one of the widespread diseases,especially for the long-term sedentary,elderly,accidental injury and other groups has a very high rate of causing paraplegics,seriously affecting patients’ daily life.Studies have shown that effective cervical exercise can help some patients with severe cervical vertebra to achieve rehabilitation.However,cervical exercise is a long-term and moderate way of recovery,and it is difficult for patients to exercise regularly and at an appropriate angle.Using intelligent monitoring technology to identify patients’ cervical exercise movements in real time,replacing medical staff to control patients’ cervical exercise bending degree and urge patients to carry out rehabilitation training,which is of great significance for patients to carry out long-term exercise independently.In view of the above requirements,this paper mainly studies the head posture recognition in the rehabilitation exercise scene,and the specific work is as follows:(1)A new head pose recognition method based on dual feature extraction is proposed.First,face detection is carried out for all faces in the image data set,and the coordinate of the image with the highest confidence is obtained by comparing the results of the detection,so as to complete the cropping of the target face.In order to further reduce noise interference,LBP features are used to counter the interference of external illumination and other factors,and facial features are pre-extracted.Then,the network based on regression and feature aggregation is used to refine the model.Before clustering,the model learns important parts by means of spatial grouping,so as to realize the fine-grained structure mapping of the model.Finally,the Euler angles on the three head angles are obtained.Finally,the validity of the algorithm is verified by experiments on the public data sets AFLW2000 and CAS-PEAL-RI.(2)In order to realize real-time synchronous analysis of the head posture in the video flow through the monocular camera,an online posture recognition method for cervical spine rehabilitation exercise based on the monitoring video stream was proposed on the basis of fine-grained structure network,and the running rate was about l0FPS.This method realizes online face detection through SSD algorithm,and then realizes the target face selection in video stream by confidence comparison,and inputs the target face into the trained network model in order to obtain the pose of the current frame.In order to evaluate the exercise quality of patients with cervical spondylosis,it is necessary to divide the activity interval of patients’ cervical vertebrae.The exercise interval is divided into "tricolor interval",which includes:safety interval,warning interval and danger interval.Secondly,due to the continuity of head posture,a marker position method was designed to count The Times of patients’daily cervical exercise to each interval.According to these statistical data,the exercise quality of patients with cervical spine was finally evaluated.Finally,the front-end page is developed by using the front-end framework vue to observe the data,and the visualization of the data is completed.