Video based action detection and recognition human using optical flow and SVM classifier

Computer Vision is gaining importance with wide applications in video surveillance, video retrieval and analysis, and human - computer interaction. Detection and recognition of human action from the video databases is really a difficult and challenging task. In this paper, video based human action detection and recognition is addressed and performed on KTH dataset and on real-time videos. At first, hundred frames are extracted from each video sequence and optical flow between the frames is computed. The data extracted using optical flow is converted to binary image. Then Histogram of Oriented Gradient (HOG) descriptor is used to extract feature vector from the binary images. These feature vectors are given as training features to Support Vector Machine (SVM) classifier to prepare a trained model. For testing, we created our own real time videos which consist of actions: walking, jogging, running, boxing, hand waving and handclapping. For each video the same kind of features are extracted and given to SVM classifier for classification of actions.