Physical Abuse Activity Recognition using Landmarks of Human Posture and DTW Classifier

Video Surveillance analysis can help to recognize the different types of crime activities i.e., abuse, fighting, robbery, shooting, etc. The main purpose of this study is to recognize physical abuse activities like kicking, dragging, and beating along with their postures. This study uses the values of landmarks (Left_Wrist, Right_Wrist, Left_Hip, Right_Hip, Left_Ankle, Right_Ankle) of human postures in video frames. Euclidean distance (D<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</inf> to D<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">10</inf>) between the landmark vectors is used as the input for DTW (Dynamic Time Warping) classifier. DTW measures the similarity between training vectors and test vectors to recognize the type of physically abusive activity and classify the posture. The UCF-101 Crime public dataset is used. The empirical analysis indicates a 96% accuracy.