Video Based Human Activity Detection, Recognition and Classification of actions using SVM

Abstract

Human motion analysis which includes activity detection and action recognition is currently gaining attention from computer vision researchers. Automatic monitoring of human activities and actions using computers has found significant applications in video surveillance, monitoring of patients and sports applications. With the tremendous advancement and development in digital video libraries, automatic interpretation of videos will save human effort in analysis and interpretation. This has led to the development of robust techniques in the field of computer vision. Human activity detection and recognition includes detection of human, tracking of human and recognition of actions. In this paper, detection of human is done using Gaussian Mixture Model, tracking is done using optical flow, recognition and classification of actions is done using SVM Classifier. The experiment is carried out with two public datasets KTH and Weizmann which are the videos with constant background. The classification accuracy for KTH dataset is 92.48% and for Weizmann dataset the classification accuracy is 93.64%.