Several Studies Video Behavior Analysis

Behavior analysis in the video is a popular research area in the filed of computervision nowadays. Recently, many problems are released and studied worldwide forhelping the process of industrializing it. Especially, Microsoft has launched its newproduct, named Kinect, by using the behavior analysis technology. And there is nodoubt that, research in this area will be more focused than ever. Many difficult problemsare still existing in this field, especially when the environmental factors are considered.This greatly motivates our research in this paper.Two important problems in behavior recognition are considered in this paper.Basically, our recognition framework is based on the segmentation of object area(people in this paper). Hence, we should first study how to detect and localize personsin the image. Then, we can use this localized area to finish behavior analysis.In the process of detection, we first use motion information to help reduce thedetection area. Then, we only need to carry out detection algorithms in this candidatearea. In this paper, we propose a new method to localize people in the image. Ourmethod is based on Hough voting, which transforms the original image into a parameterspace. Different from previous Hough-voting based algorithms, our method novellyconsiders the correlation within the sampled features. Final experiments prove that itcan increase information in the process of feature description.The behavior recognition is carried out on the candidate area. In this paper, wepropose a novel hierarchical framework to recognize human behaviors in video. Afterthe preprocessing step, we obtain the human silhouette image and its correspondinghuman body image for further analysis. An improved approach is provided to finish thehuman posture representation, which considers the human body as a pictorial structureand makes full use of information of body parts. This pattern feature representation canbe well suited to highly articulated postures without losing recognition efficiency.Hidden Markov Model (HMM) is used to model the human action sequence which iscomposed of various prototypical postures changing over time.Sufficient experiments are carried out at each end of alogorithms sections, whichare used to verify the proposed method and do valuable comparisons with previoussolutions.