Study of efficient human detection and tracking algorithm for surveillance

Abstract

Human tracking is one of the crucial tasks in intelligent video newlineanalysis and has wide applications in surveillance, homeland security, humancomputer newlineinteraction, Advanced Driver Assistance System (ADAS), robotics, and so on. Thus, a great deal of literature on the development of human detection and tracking algorithms report promising results under various newlinescenarios. However, detecting and tracking the human is challenging aspect newlinedue to the following issues: significant pose variations, illumination variations, newlinescale variations, cluttered background, and occlusions. The higher hardware newlineresource utilization is yet another issue in real-time human detection systems. newlineThe main objective of this research is to develop the efficient human detection and tracking algorithm for surveillance. In order to meet this objective, this research work addresses feature extraction algorithms for human newlinedetection and effective appearance model based on a sparse representation newlineframework for human tracking. The edge and Histogram of Oriented Gradients newline(HOG) are considered as features. These features encode image regions as high newlinedimensional feature vectors and provide high accuracy for human detection. newlineThe edge features are extracted using the proposed block-based Canny edge detection algorithm. This algorithm characterizes the block of an image into uniform, texture, medium edge, and strong edge. The threshold newlinevalues are found using block statistics rather than whole image statistics in newlineorder to extract the edges under illumination variation and noise. The proposed newlineblock-based Canny edge detection algorithm newline newline