Design of Mobile Application System for Recognition and Analysis of Dynamic Running Posture

Aiming at sports injuries caused by incorrect running postures, this paper designs a mobile application system based on Android for the recognition and analysis of running postures. According to the profile running video, the system extracts the frame set of the time sequence from the filtered images, using the OpenPose algorithm to achieve runners’ 18 joint coordinates and calculate the corresponding eight bone vectors and four joint angles. A dynamic skeleton with four joint angle time sequences is constructed by convolutional neural network fitting and using the dynamic time wrapping (DTW) algorithm to calculate the similarity between the measured and the standard dynamic skeleton, finally identifying the time when the abnormal running position appeared and the corresponding body position. All user information is written into the MySQL database as JSON data, and the front-end interaction uses Android Studio to present visual running posture analysis data. The results show that the system can identify a variety of abnormal running postures, can better adapt to complex backgrounds, has a higher recognition accuracy and calculation rate, and provides users with scientific real-time movement guidance.