Identification Of Human Body Parts Based On Deep Information

Human gesture recognition and motion analysis are important research directions in the field of computer vision.Identification of human body parts can lay the foundation for human gesture recognition.In recent years,the rapid development of depth image technology has provided new methods for identification of human body parts and gesture recognition.The depth image directly reflects the three-dimensional space information of the objects in the scene.And it is not affected by the light source irradiation direction,color,texture,and the reflective properties of the surface of the object,compared with the traditional visible light image,it has incomparable advantages.The depth image can greatly simplify the complex tasks in the field of computer vision such as target recognition and 3D reconstruction.This paper focuses on the identification of human body parts based on depth image information.Inspired by a pixel classification method based on depth comparison features proposed by Shotton from Microsoft Cambridge Research Institute,we built a human depth image data set,based on the traditional depth comparison feature,we use random forest training sample data to obtain human body parts classification model.By analyzing the problems of traditional depth comparison feature in human body parts recognition,we improve the depth comparison feature and propose the central circle offset depth comparison feature.We use the center circle offset depth comparison feature and random forest for human body recognition studies,we analyze and compare each factor that affects the classification result to get the best possible classification model.Human body parts recognition results on the test data set indicate that compared to the traditional depth comparison feature,with no increase in calculations,the average accuracy of human body parts recognition based on the central circle offset depth comparison feature increased by 4.92 percentage points.Based on the central circle offset depth comparison feature,we can not only extract the human joint points more accurately and generate human skeleton information,but also can obtain a human body part classification model with better generalization ability under the condition of small data sets.