Data-driven 3D Human Motion Posture Tracking And Application Research

At present,the real-time capture of three-dimensional human posture in games,the mainstream method uses Kinect equipment to capture human posture,and with the help of XBOX,the incoming data is analyzed and processed.The advantage is that it has a depth camera that can generate more accurate three-dimensional bone data,but the disadvantages are also obvious,the equipment is expensive,and it is restricted by light and angle,resulting in Kinect can only be used in indoor scenes that meet the requirements.In order to solve the above problems,this article is based on the lightweight convolutional neural network to quickly obtain bone data,and improved on the linear Kalman filter.The main work of this paper is divided into the following points:1.Use a monocular camera to collect images,and optimize the convolutional neural network to make it faster than other convolutional neural networks with more layers.2.Use deep neural networks to capture the posture of human bones.For bones with higher degrees of freedom,according to the advantages and disadvantages of different representation methods of bone data,this article will use the quaternion method to transfer the bone information in Unity3 D.3.The resulting bone data flash problem.Kalman filtering is improved according to the law of human motion posture.The collected 3D bone data is tracked using improved Kalman filtering.Finally,in order to verify the effect of the improved Kalman filter,a music game system was developed on the basis of the Unity3 d game engine.Standing within the range that the monocular camera can capture,the three-dimensional cartoon character on the screen will make a mirror image.The purpose is to make the three-dimensional virtual character collide with the flying objects to get points.Using the game made to verify the improved Kalman filter,the effect is better than that of the pre-improved Kalman filter.