Research On Indoor Human Behavior Recognition Using Kinect

Human behavior recognition technology is an important field of machine vision, and has challenging and attractive. Its application areas are mainly concentrated in the intelligent video monitoring(Such as hospital, traffic, home, school, military training, athletic field, etc), human-computer interaction, virtual reality, some behavior recognition less areas involved, like content-based video retrieval and intelligent image compression in the Internet, and so on. Therefore, from the point of broad application prospect of behavior recognition technology, it has huge and potential economic value and social value. However, from the current technology of human behavior, the defects are as follows:(1) Because the environment adaptability is low and non humans and the dynamic object illumination is excluded difficulty, it causes the interference of identification;(2) The human toward angle adaptability is not strong,(3) The recognition rate is not high.To solve these problems, a method of indoor human behavior recognition via a Kinect application is proposed. Based on the Microsoft Kinect device researching recognition of human behavior on the small range of indoor conditions, we get RGBD datas, research the Kinect platform and its data, and the recognition and behavior of the platform. The multi feature fusion of human skeleton of 3D feature and HOG feature representation method is proposed, which overcomes the defects of the technology(1) and(2) points, and also solves the instability of Kinect data acquisition. Then online dictionary learning and sparse principal component analysis are used to the characteristics. Finally combined with multi task large margin nearest neighbor(MT-LMNN) and linear support vector machine(LSVM), the behavior is classified based on the scoring mechanism to make the best decision, effectively overcomes the defects of the technology(3) points.Through a large number of experimental data of the test, and compared with the advanced algorithm, the results of this paper show that the feature fusion and ensemble classifier with the scoring mechanism used in this paper has better recognition rate and treatment rate. This research has a good reference for the further research, and also has certain academic value and application prospect.