Research And Implementation Of The Human Body Pose Estimation System Based On A Statistical Model

Human pose estimation is a process that using video or sensing technologyto record human body movement, and to obtain motion parameters and bodystructure for the posture analysis. Human pose estimation was applied widely tohuman-computer interaction, sports training, healthcare, animation and television,and many other fields. Pose estimation based on sensor technology especially hasbroad application prospects and huge economic and social value. Althoughresearches on video-based human pose estimation have achieved good results,but it is inherently influenced by the environment, computationally intensive anda invasion of privacy which make its applicable scope limited.Compared to the video based pose estimation, sensor-based estimationmethod is not environmental restricted, easy to operate and does not infringe onprivacy. This article will use inertial measurement units to gather information ofhuman body movement, and a deep study on inertial measurement-based humanpose estimation algorithm and fall monitoring algorithm will be conducted, andfinally using of Bayesian networks for the forecast and estimation on humanbody posture were explored.In this article, the human motion information was collected by sensors, thepose estimation algorithm includes the calibration between sensors and humanbody parts, the ground contact detection and the segmental kinematics. Besides,a research on the warning of falling down was conducted based on theacceleration characteristics extracted from the motion information collected bythe sensors. The the algorithm was verified by experiments, it showed that thealgorithm can estimate human body pose and at the same time conduct earlywarning of falling, with good results.This paper studied the problem of human pose estimation through BayesianNetwork which is a graph model. And we built the dynamic Bayesian networkmodel of human body walking, combined with the information from the sensorsas the observation information, the pose estimation and prediction was conductedthrough the inference functions of the network, a long-distance walkingexperiments showed the good performance of the algorithm.