Human Pose Estimation Using Depth Information

One of the key directions of computer vision research is human body estimation.Human body estimation has a wide range of application scenarios,such as medical aid diagnosis,human-computer interaction system,and intelligent monitoring system.Since ordinary optical images are easily interfered by external environments such as light and shadows,and the occlusion of human body parts,as well as the variety of human body shapes and clothing,it is a very complicated problem to recognize human body estimation from optical images.Therefore,this paper uses the depth sensing device Xtion to collect the depth image,and applies the DenseNet module based on the Generative Adversarial network to learn the key joint points of the human body and judge the behavior of human.In the last,we apply the technology to the scene of intelligent care for the elderly.The main research work in this paper focuses on the detection of human joint points in depth images.The main contributions of this paper include the following aspects:1.An DenseNet module based on the improved Generative Adversarial network model is introduced for human body estimation for the first time,which is beneficial to solve the problem of highly nonlinear mapping caused by obtaining the 3D coordinates of human joint points from the depth image.The experimental results in the public data set show that the key joint detection of the human body proposed in this paper has obtained the best detection result in the current published algorithm,which is 4% higher than the best result.2.An end-to-end method for reconstructing the 3D pose of the human body from depth image restoration is proposed.The use of depth images effectively solves external environmental problems such as illumination and shadows,and the obtained 3D results effectively solve human occlusion problems.The end-to-end method makes human body estimation more convenient for various scenarios.3.Our work completed the elderly intelligent care system with depth images and gave the early warning of fall pose of the elderly.After obtaining 3D skeleton map,we analysis several neighboring skeleton maps,the height and speed of human key point are used to judge the posture of the elderly.In the actual test environment,the accuracy of the fall judgment of the system was found to be above 90%.