Recognition Of Humans Inside Video

Computer Vision broadly refers to the discipline where extraction of useful 2D and/or 3D information from one or more images is of interest. Since the human visual system works by extracting information from the images formed on the retina of the eye, developments in computer vision are inevitably compared to the abilities of the human vision system. One of the basic tasks of the human visual system is to recognize humans and objects and spatial relationships among them. Similarly, one of the main goals of computer vision researchers is to develop methods for localization and recognition of objects in a scene. A special case of this general problem is the recognition of humans and their activities. In this monograph, we outline some of the methods that have been recently developed towards achieving this goal, and describe our present research in this area. Specifically, we divide the recognition problem in two parts-human identification using face and fait, and human activity recognition.The basic input for recognition systems is a video sequence (a single image being a special case of it). Unless otherwise specified, we will assume throughout that the input data lies in the visible spectrum. However, video is collected under different conditions and it is unrealistic to expect the same kind of performance even as the resolution of the images varies. Face recognition algorithms usually require high resolution video where the face is the predominant subject in each frame. Gait recognition can work with lower resolution data provided one can extract the motion of the different parts of the human body. Our experience is that video data on which gait recognition algorithms perform reasonably well, usually do not have the resolution for satisfactory performance with face recognition algorithms. In many practical situations, the activity being performed is often of more importance that the identity of the person involved. Also, the input data may be of such a resolution that it is impossible to identify any single individual, but it may be possible to understand the tasks being carried out by a group of people. Under such circumstance, activity recognition algorithms are required.Face recognition under unconstrained condition is a difficult problem in face recognition area. Two important steps in designing face recognition algorithm are feature extraction and classification algorithm designing. The prior face recognition algorithms have concentrated on one of the two. However, the method not only has an optimal feature extraction algorithm but also has an optimal classifier is not yet available. Toward this end, Via reuse of the PCA(principal component analysis)concept for all the major processing stages, we propose a new video face recognition algorithm, which contains face detection,2D-to-3D face modeling, bit-plane feature fusion and self-PCA recognition. Experimental results illustrate that the proposed algorithm outperforms the selected benchmarks representative of existing techniques.Recently, much research has been concentrated on video-based face recognition. The still image problem has several inherent disadvantages. In this paper, Due to the abundance of frames in a video, the problem of face recognition inside video is explored, and a novel post-processing HMM-based solution is proposed to improve the robustness of the recognizer under unconstrained environment. Experimental results have impressively indicated the effectiveness of the proposed post-processing solution to tackle the problem of interferences.The quality of human silhouettes has a direct effect on gait recognition performance. This paper proposes a robust gait representation scheme to suppress the influence of silhouette incompleteness. By means of dividing human body area in a video sequence into several sub-areas, representing each sub-area through an ellipse whose parameters can be calculated from the corresponding motion information extracted from optical flow field, a new body structure model called multi-linked ellipse model is established. In the recognition stage, the parameters of model are finally used to achieve gait recognition based on dynamic time warping technology. Experimental results prove the higher performance of the method.