Deep learning real-time detection and correction system for stroke rehabilitation posture

Abstract

Cerebrovascular diseases are becoming younger, and rehabilitation treatment becomes the importance of future recovery. Some rural areas have limited medical re-sources or people living in more remote mountainous areas have certain difficulties to receive treatment in hospitals. To solve this problem to build a set of OpenPose based human posture recognition algorithms to apply to stroke rehabilitation, to process the action of the characters in the image. Use the frame extraction algorithm of motion capture data to obtain the key frames of the character's movement, so as to analyze the details of the human movement .Then identify the body movements, carry out the stroke rehabilitation OpenPose human body model training, and use the Law of Cosines to calculate the angles of 3 points in BODY25, so as to analyze the details of the body movements to identify the body movements and provide the correct posture to guide the rehabilitation patients to the posture angle Rehabilitation.The experimental results show that the test subjects have a correct rate of posture and movement of 53.3% before using the system and 97% after using the system, and an improvement rate of 43.7%.