Improved Multi-feature Computer Vision for Video Surveillance

Abstract

Computer vision deals with automatic extraction, analysis, and understanding of information from image or video. Recent research in the area of computer vision mainly focused on developing intelligent systems for detecting and observing human behaviors. Human detection is the process of locating human automatically in an image or video sequence. This paper presents an efficient intelligent surveillance system using features, namely HAAR, local binary patterns, and histogram of oriented gradients to detect the presence of human being. Support vector machine is used to train the system. In most of the work in computer vision, researchers used different datasets and different features. This paper compares the result of the mentioned features using a common high-definition dataset. To reduce the time complexity of high-definition video, the proposed method normalized the extracted video frames into equal size. The experiments are performed on images and videos and the performance of the system is presented term of true positive rate, false positive rate, accuracy, and computation time.