Personnel Positioning And Motion Analvsis In Virtual Arrangement System

Under the positive influence of the rapid development of deep learning,computer vision technology has gained rapid development momentum in recent years.More and more fields introduce computer vision,through more and more mature technology to help people save a lot of human and material resources,even in some specific areas to replace human to complete the work that human beings can not complete.With the great success of other industries such as security,sports and performance industries have gradually introduced computer vision technology.Through the camera to capture the action picture of athletes or actors,analyze the action standard degree of athletes or actors,assist athletes or actors to correct the action,so as to obtain better training effect.After carefully consulting and reading a large number of relevant domestic and foreign literature,this paper summarizes the advantages and disadvantages of various methods and applicable scenarios.On this basis,aiming at the large and medium-sized stage group performance scene,the target tracking and positioning and movement analysis in multi person scene are studied in depth.In multi-target tracking and positioning,several traditional tracking methods and detection methods based on deep learning are studied and analyzed.Combined with the needs of actor's action analysis,the actor's skeleton is extracted by bottom-up method,and the actor's joint point information is obtained.The caffenet network model is used to detect and identify the self-designed identity objects,so as to achieve the matching of skeleton and actor in multi-objective scene.For actor positioning,triangle positioning method is adopted:firstly,Zhang Zhengyou's chessboard calibration method is used to calibrate the camera,and the internal and external parameters of the two cameras are solved according to the self-defined world coordinates and the image coordinates in the imaging of the two cameras.Finally,the actor's world coordinates under the self-defined world coordinate axis are solved through the polar geometric constraint relationship.In multi-target detection,it is inevitable that the target will be occluded and lost.Therefore,this paper proposes a prediction and tracking method based on Kalman filter and motion track fusion.For the action analysis of actors in the scene,this paper first makes affine transformation to the detected skeleton to reduce the error caused by the camera angle and the distance between actors and the camera.Then this paper extracts the angle features of skeleton joints to segment the action sequence,then uses the dynamic time warping method to align the segmented sequence with the standard template sequence,calculates the distance between them,maps the distance to the scoring space to evaluate the actor's action.