Virtual Self Defense Trainer - Analyzing and Scoring User Pose

Self-defence is an important skill for self safety especially for women in times of danger. However, self-defence training is not accessible to many. This paper presents a framework that estimates human pose using a webcam and openpose and analyzes the pose in order to give a score and performance based review. This framework aims at teaching self defense to users. The idea is to capture the pose of the user via an RGB camera and estimate pose using openpose. Then features are extracted and the poses aligned in the temporal domain. The result of the overall pipeline is a score and natural language feedback for each feature extracted. The results show that our system performs quite well on the video dataset prepared.