Research And Development Of Home Motion Guidance System Based On Attitude Estimation Algorithm

In recent years,with the rising enthusiasm for sports,sports platforms and APPs have emerged,allowing athletes to perform targeted physical exercise at home.According to the survey,when exercising at home,the movement guidance and correction of the athletes is a key part.Traditional exercise assistance relies more on human judgment;or requires the human body to carry equipment to collect exercise data,which is easy to cause inconvenience in exercise.With the development of computer vision,the use of deep learning methods can make motion assistance systems more intelligent.Aiming at the application scenario of home exercise,this paper makes a lightweight improvement on the Open Pose algorithm for attitude estimation,proposes an action comparison method based on Dynamic Time Warping(DTW),and completes the development of a home exercise guidance system.In the detection of human key points based on deep learning,the Open Pose algorithm has high accuracy and good real-time performance,but it also has the problems of high complexity and high consumption of computing resources,which is not conducive to application.This paper proposes a lightweight improvement for the Open Pose algorithm to obtain a lightweight human key point detection algorithm with both accuracy and real-time performance.Based on the Open Pose algorithm,firstly,the lightweight Mobile Net v3 combined with the attention mechanism is used to replace the original Open Pose backbone network;further,the number of prediction branches of the original algorithm is reduced,and the small convolution cascade is used to replace the large convolution,thus obtaining an improved Lightweight Algorithms.Finally,the experimental comparison of the algorithm proposed in this paper shows that under the premise that the detection accuracy is almost the same,the algorithm proposed in this paper greatly reduces the computational consumption,and the video detection rate can reach 25 fps,which is about 16 fps higher than Open Pose.In order to better evaluate the movement situation of the athlete,the DTW algorithm is used for action comparison,and a DTW-based action segmentation method is proposed.First,the coordinates of the key points of the human body are obtained through the pose estimation algorithm,and the coordinates are converted into joint angles.Further,combined with the joint angle information,a standard action template is made in a non-specific way,and the burr mutation generated in the acquisition process is preprocessed.After completing these steps,perform motion segmentation and alignment.First,set the detection standard template of the start and end of the motion,and perform the start and end frame detection in the form of sliding window.The sequence in the window is compared with the detection template by DTW,and the minimum value of the comparison result is determined as the start or end bit,and finally,the sequence between the start frame and the end frame is further compared and analyzed.Experiments verify the effectiveness of the above processing alignment and the accuracy of the segmentation method.Finally,the design of the home motion assistance system is completed.The system designed and implemented in this paper has basic functions such as user registration,login,and password modification.It also has the function of voice prompts during exercise and viewing exercise data.The functional modules of the system are tested and all can meet the expectations.