Performance Analysis of Spatio-temporal Human Detected Keyframe Extraction

Abstract

Closed circuit television (CCTV) surveillance for detecting the humans involves an expanded research analysis especially for crime scene detection due to various restraints such as crowded annotation, night footages, and rainy (noisy) clips. The main visualization of the crime scene is to recognize the person in particular obtained in all frames is a challenging task. For this occurrence, Content-Based Video Retrieval (CBVR) method refines the collection of these video frames resulting keyframes to reduce the burden of huge storage. Here, Spatio-Temporal classifiers method as an added advantage with frame differencing and edge detection method reports the human detected keyframes without the termination of background regions in order to negotiate the crime scene more efficiently.  The main objective of this paper is to analyze the obtained keyframes with Human detection pointing a distinctive between Spatio-Temporal HOG-SVM and HAAR-like classifier to survey the optimum. Finally, the resulting keyframes mutated with the canny edge detection method by HOG-SVM sequel with greater accuracy level of 98.21% compared to HAAR-like classifier.