Sportstraining System Research Based On The Human Body Posture Estimation Algorithm

With the rapid development of computer technology,people began to use virtual reality technology to realize a scientific and quantifiable sports training scheme,in order to cast off the state that traditional sports training completely relied on the experience.The main task of the human body pose estimation is to obtain the position and the angle of all human body parts in the 2D or 3D space.By establishing the mapping relationship between the human body features and the human body postures in the image I achieved the purpose of estimating the pose parameters.This paper established a human assistant training system,and applied it to the golf sports training.Detect and track the movement of human body in the video in order to achieve the purpose of analyzing movement posture and providing training proposal.Realize the transformation of human motion training method based on the experience to the method based on computer vision technology.The main work and contributions of this paper include:1.Study the basic principles of three-dimensional sensor Kinect.Kinect has three cameras.I can finish the collection work of the sequence color image and its depth image information by using the Kinect,the image is used as the input of the later sample.Kinect can also track the human skeleton,we use its plugin OpenNI to achieve automatic extraction of 3D data of 15 joints of the human body,and finally we complete the accuracy verification by comparing the pose data obtained manually with the data obtained by the Kinect 3D sensor.2.Study the common method of image feature extraction and human target detection.The traditional target detection solutions areless robust for the dynamic scenes,the influence of the background shadow disturbance and illumination changes is relatively obvious,so it is crucial to establish an adaptive strong and robust background model for target detection.This paper proposeda background modeling algorithm based on ViBe,and use neighborhood pixels to create a background model.The experimental results show that this method can overcome the shortcomings of the traditional method,and also can detect and extract the contour of human body from the video precisely.3.Study human pose estimation algorithms based on model and non-model.Firstly propose a human pose estimation algorithm based on contour features and image processing.Then carry out the edge detection to the binary contour image obtained by the target detection.Finally use the image processing methods,such as horizontal scanning,human body length scale constraint to realize the human joints pose estimation.The experiments show that this method can extract the body's major joints data more accurately.4.Take golf as an example,we can use the human pose estimation method to get the swing posture parameters both of trainers and coaches,and build anassistant training system.The system takes the joint angle trajectory and attitude similarity as two auxiliary indicators,and compares the joint angle trajectory of the trainers and coaches to conduct analysis and guidance according to the attitude similarity.