Research On Non-rigid Point Set Registration Method By Global And Local Mixture Structure And Its Application

The purpose of point set registration is to obtains the point-to-point correspondence between two or more point sets,and solves the space transformation between these point sets according to the correspondence.As one of the basic components in the field of computer vision and pattern recognition,point set registration is widely used in image processing,object detection and 3D construction.Therefore,point set registration has extremely important theoretical research significance and practical application value.Firstly,this paper expounds the research status of point set registration at home and abroad,analyzes the defects in all kinds of point set registration methods and the shortcomings in practical application.Then,aiming at the disadvantages of geometric structure based non-rigid point set registration method,a non-rigid point set registration method based on global and local mixture structure feature is proposed.And then,aiming at the problem of track-to-track association under sensor bias,a track-to-track association method based on the global and local mixture feature is proposed by using point set registration method.Finally,aiming at the problem of multi-object tracking in the perspective of UAV in the presence of violent random movement of UAV,a multi-object tracking method of UAV based on robust detection-to-track association is proposed,which provides some reference for the application of point set registration method in data association field.The main research contents of this paper are as follows:1.A non-rigid point set registration method based on global and local hybrid structure features is proposed.In this paper,two kinds of feature descriptors,which are mixed with Euclidean-distance and Angle,are proposed to obtain the global and local geometric structure feature differences of the point set,and the the two feature differences is used to construct the cost function,thus the problem of solving the correspondence of the point set is transformed into a linear assignment problem.Then the thin plate spline function is used to construct the space transformation model of the point set.Finally,simulation point set registration experiments and real image registration experiments are used to verify the effectiveness of the proposed registration method.2.A track-to-track association method based on global and local mixture feature is proposed.To solve the track-to-track association problem of distributed sensor system with the sensor bias,using the Angle and Euclidean-distance measuring the global and local geometric structure differences between two tracks,and building the registration object function based on mixture structure differences,will track-to-track association problem is converted into a distribution.Finally,the validity of the proposed correlation method is verified by simulation track correlation experiment and real data set experiment.3.A multi-object tracking method for UAV based on robust detection-to-track correspondence is proposed.Aiming at the multi-object tracking problem of UAV with large deviation between detection and track due to random movement,a robust detection-to-track association algorithm is constructed based on the idea of point set registration.The algorithm used Euclidean-distance,angle and color to measure the geometric structure feature difference and color feature difference between track and detection.Then the three differences are combined to construct the cost matrix,and combined with the thin plate spline function to establish the registration-association integration object function.Then the track management was used to realize the multi-object tracking.Finally,the effectiveness of the proposed tracking method is verified by the tracking experiment on the data set collected by UAV.In general,the performance of the non-rigid point set registration method proposed in this paper is superior to the most classical point set registration methods at present.The proposed registration method is applied to track-to-track association of distributed sensor system and multi-object tracking from the perspective of UAV,which provides more references for the application of point set registration method in the field of intelligent transportation system,and provides certain theoretical and technical support for the development of intelligent transportation system.