Video Keyframe Extraction Based on Human Motion Detection

Abstract

AbstractDue to the substantial growth of CCTV surveillance data, it is very hard to cumulate the crime scene information from a long durable video collection as frames. Keyframe extraction is used to eradicate the non-essential frames in order to reduce the processing time of an entire video. Still, keyframe extraction lags to gain more accuracy on determining the crime scene with human detection, thus the spatiotemporal feature extraction approaches the human motion detection phase using the HOG descriptor along with the SVM classifier was reviewed from the existing methods. In this study, two methods are implemented by a combination of frame difference method with HOG along SVM on various edge detection methods, predicts the optimization of human motion detected keyframes. These extracted human detected keyframes are sustaining the local features as keyframes for depicting the crime scene as a clear summarized report. Finally, the experimental result shows that spatiotemporal feature extracted keyframe through Canny edge detection achieves 98.73% as recognition accuracy.KeywordsCCTV surveillanceKeyframe extractionHuman motion detectionSpatiotemporal feature extractionEdge detectionHistogram of oriented gradients (HOG) descriptorSupport vector machine (SVM)