Behavior recognition in surveillance video using temporal features

Manual video surveillance is highly expensive and inconvenient in continuous monitoring by security personnel. So automatic video surveillance is needed. A large amount of security measure is required in public and private sectors due to terrorist activities. In this paper, an automatic activity recognition approach is proposed. The difference image is used to extract the motion information based on Region of Interest (ROI). The experiments are carried out on the Weizmann dataset considering four activities viz (walking, running, waving one hand and waving both hands) with multiclass-SVM scheme for classification. Experimental results shows an overall performance of 92% to recognize performed actions. The proposed method demonstrates good results on Weizmann dataset and results are comparable with well-known existing methods.