Hum A N Action Recognition System Based On RGBD Data

Human action recognition has been the research focus in the field of computer vision and it has been widely used in intelligent monitoring and so on. Traditional research was based on RGB video. In recent years, with the Microsoft Kinect and other depth cameras proposed, human behavior recognition based on RGBD data get more and more attention of academics and industry, RGBD is short for RGB-Depth. Because of these reasons mentioned above, Human behavior recognition based on RGBD data has very high research value.This paper mainly research on Human behavior recognition based on RGBD data, proposed a real-time human action recognition scheme which based on multi skeleton feature and implemented a human action recognition system. This paper mainly achieved the following three tasks:Firstly, in the view of the existing some 3d human body skeleton characteristics of deficiencies, this paper proposes a feature extraction method of three dimensional human body skeleton; Secondly, by using the proposed features with existing fusion, make up the shortfall, puts forward a real time recognition method which fusion multi 3d human body skeleton features; Thirdly, this paper implements a RGBD data acquisition and human behavior recognition system, which can gathered multi modal data through the device, and the collected data is used to identify the human behavior, the recognition algorithm combine the work mentioned above and a depth image feature, then use SVM to classify action. The feature we proposed in this paper has many advantages above other existed features. The algorithm proposed in this paper is fast and has high accuracy. The system implemented in this paper is robust.