Research On Human And Face Detection In Security Video

Security is undoubtedly extremely important for everyone, artificial intelligence more and more widely be used for people’s lives to ensure the safety of people’s lives and property. Because intelligence tools can detect the emergence of strangers quickly and accurately and efficiently. This paper is based on this purpose. Research it’s possible that added in the intelligent security of the human body detection and face detection algorithm, the video stream captured by camera can be process in real-time. For this purpose the main contents of this paper can be divided into three parts:1. In this paper, the video stream is taken directly from the camera for processing, if there are moving target, it will start work that detect human body and face. Frame difference method generally used to detect whether there are moving targets. For empty frame difference appears after and because of poor video quality, lots of noise, multi-frame fusion of the proposed method. Moving object appearing in a video stream, you can follow up with human detection and face detection.2. Common human detection algorithm and improved algorithm under certain conditions. In the moving object is first detected whether the human body, its purpose is to prevent criminals face obscured face detection leads to failure. Since the characteristics of HOG feature extraction algorithm, making it subject to environmental factors and other effects of light and little physical activity, can be described to be detected moving target. Then use the human body through the SVM algorithm determined after training for HOG feature extraction determination. Where the lack of practical application of the training data, resulting in the classification effect is limited, based on positive feedback and proposed SVM classifier model that can take full advantage of the results of the determination to train the SVM classifier.3. It is divided into two parts to achieve the required face detection goal, the first, face detection, which has been determined to be a moving target in the detection of human face image, this detection algorithm similar with human detection, include the feature value extraction, and the characteristic values determined. Usually extracted Haar-like feature because Haar-like feature can be a good representative of face image, then use Adaboost algorithm to discriminate. It was carried out face detection in image of detect human body, the method can be reduced to a large extent the face detection window to be picked up. Thus indirectly reducing the amount of computation, and improve the face detective of performance.