Human Motion Detection and Recognition from Video Surveillance Based on Machine Learning Approaches

Abstract

Detecting human and human motion through a video surveillance system is an important task. It helps in various fields, like crowd analysis, identifying abnormalities in human behavior in public places and identifying a person and their gender. To detect a human and its motion in a video, the first step is to detect the moving object correctly. Background subtraction method is an efficient way to detect the moving object efficiently on the foreground frame. Object detection and extraction could be performed in various ways. Once the object is found, then a classification method is applied to recognize the object and its motion. In this paper, first, the background subtraction method is applied to detect the person in a video input file. Then, this method is applied to track the human’s motion through the entire video. In the second step, the histogram of oriented gradient method (HOG) is used to extract the features from the input video file. And, in the last step, support vector machine (SVM) is used to classify the detected person’s identity and its motion. In this experiment, two different multi-class classifiers are used, SVM and decision tree (DT), to compare the performance of the classification models. Finally, a detailed comparison of each person and the motion class is performed to compare the classification rate of the two classifiers.