Research And Application Of Behavior Recognition Technology Based On Kinect

In recent years,due to the behavior recognition technology can effectively improve the sensing capability of the system,it has been widely used in the field of human-computer interaction,intelligent monitoring and virtual reality,so the study of human behavior recognition received more and more attention.However,the diversity of the real environment,the changes of light,clothing,complex background make the human behavior recognition become a very challenging research field.With the Microsoft Kinect depth camera release,depth map directly reflects the three-dimensional characteristics of the surface,to overcome the traditional image recognition difficulties,the researchers used different research methods and ideas,have obtained the certain research results.This paper uses the Kinect sensor to obtain the moving human color,depth image and bone information on key technologies related to human behavior recognition,behavior feature extraction and expression,and then complete the understanding and recognition of human daily action.The main work of this paper is as follows:In the aspect of feature extraction and expression,expression of local features,this paper presents a gradient direction histogram based on depth image of Pyramid(PHOD)behavior expression method,and further discusses the recognition based on RGB image and depth image feature and the two feature images after feature fusion in series using SVM classifier training after.Based on the characteristics of the human body structure,the Kinect sensor is used to obtain the information of the human bone joints,and the angle between the joints and the modulus ratio are calculated to describe the human behavior.Behavioral recognition experiments were performed on challenging DHA and MSR Action3 D public behavior data sets.The results show that compared with the traditional methods,the proposed method has higher recognition accuracy.In the aspect of behavior understanding and recognition method,data preprocessing,data dimensionality reduction algorithm based on PCA to make corresponding improvements,according to the characteristics and the skeleton joints feature depth of the image is obtained by gradient descent algorithm to calculate the attribute weighting method to preprocess the behavior to improve the classification accuracy.Feature fusion,this paper proposes a feature extraction based on fusion strategy,the two-dimensional,depth image and bone information characteristics of complementary advantages,further combined with support vector machine classification algorithm for understanding and prediction of human action,the experiment results show that the proposed method is used in human behavior recognition classification has better robustness and discrimination.In practical applications,this paper presents the design and implementation of intelligent alarm function of primary care experiment system to achieve the recognition of human daily behavior and abnormal behavior,and through the system of independent recording simple daily behavior data set is feasible and effective for experimental validation of the proposed behavior recognition method.