Study Of Human Tracing And Motion Analysis Based On Computer Vision

The technology of human tracking with motion analysis is extensively used in these fields such as scene monitoring, action comprehension, posture estimation and entertainments. Actually, it is also the hot issue in the cross field of computer sight and clinical medicine. Intelligent motion recovery system developed with the technology above can analysis the person’s actions captured by its auto tracking about their motional recovery procedure to get the diagnostic data which will direct the further recovery treatment. Therefore, for the rehabilitation of disabled human, body tracking and motion analysis has important theoretical and applied research value.Vision-based3D human motion estimation can be abstracted into a complex, high-dimensional nonlinear model, due to the requirements of real-time tracking and analysis, with a certain degree of difficulty. In this paper, the single-camera tracks and analysis the scene single target’s movements. Firstly, record a video image sequence, and then create a joint body model of human. After the analysis and processing of the video image sequence, to build a body contour characteristic of human silhouette and RGB color histogram for observation model using particle filter algorithm to realize the real-time tracking of human.The characteristics of the paper are these, firstly establishing a three dimensional human model based on joint model. This model consists of6rigid bodies such as analog head, body, arms and forearms and6joints simulating necks, shoulders, waists and elbows with16degrees of freedom. The model extracts limb back bone ignoring other details to simplify the calculation and to better reflect the upper body to the requirements of the various postures. Secondly, this paper use top-down technique combined and down-top technology to estimate the initial attitude of the body, adding Haar feature based face detection technology, physical constraints and prior information as well as video of the body area characteristics, so that the track can be automatically initialized. Finally, this paper proposes a tracking model based on particle filtering algorithm. Through the motion analysis, this algorithm will establish the likelihood function of the human tracking by adding different weights of target contour, silhouette of the body, and three color channels for image together, to improve the accuracy and reliability of the matching between human projection and video image. The experiments show that the system can track human motion which parallel to the camera stable under the experimental environment.