Human Behavior Analysis And Its Application Based On Kinect

Human behavior analysis is an important research field in computer vision, its application includes intelligent video monitoring, virtual reality and various electronic equipment systems which are interact with human. Traditional human behavior analysis methods are mostly based on the color image, but there are still many difficulties in practice, such as background disturbance, the interference of environment, data information, and the feature dimension reduction algorithm for feature extraction problem have influence to human body target detection. In recent years, many researchers fusion the image of depth information and color information, and propose a lot of recognition methods. This paper uses the Microsoft Kinect as visual information acquisition sensor device combining with depth sensing technology, includes processing of color image and understanding of human behavior. In the aspect of image acquisition and preprocessing, it’s easier than traditional camera for processing. This paper focuses on the method of human behavior analysis, real-time detecting and tracking the human body in the scene. To understand the human behavior, we design a human behavior recognition system based on Kinect. In this paper, our main contributions are as follows,Firstly, this paper introduces Kinect sensor to acquire depth image, and explores the hardware structure of the device, technical specifications, and basic principle. Through constructing the experimental environment, we analyze the feasibility of human behavior through obtaining the Kinect depth image data to identify.Secondly, we study the behavior description method based on depth image. Through edge detecting, noise processing, and the classification of the target feature points to the human body and so on such visual processing methods, we distinguish the human body from the background environment. Through the study of feature extraction methods, the depth image to achieve the goal of the human body tracking, and then the basic movements of human body for identification.Thirdly, we use conditional random field model (CRF) to design a kind of human behavior analysis algorithm based on Kinect, model and recognize several simple actions. Compared with two kinds of main human behavior experiment databases, the experimental result show that under the condition of the environmental light is not stable and under the complex environment disturbance, the system always can correctly detect the human body target。 The experimental result also prove the robustness of the algorithm, human behavior recognition effect is fine.Finally, based on the human body target segmentation, the act of feature extraction and recognition algorithm, we build the Kinect experiment system and finish the recognition of certain basic human behavior in the scene.