Multiple-Target Tracking Using Spectropolarimetric Imagery.
详细信息
文摘
Detection and tracking methods are two hot research topics in the field of multiple target tracking. Often change detection and motion tracking are used to detect and track moving vehicles,but in this thesis new approaches are provided to improve these two aspects. In the detection aspect,a combined detection method is presented to improve target detection techniques. The method of combining RX Reed-Xiaoli) with change detection has demonstrated good performance in highly cluttered,dynamic ground-based scenes. In the tracking aspect,Kalman filter and Global Nearest Neighbor are applied in motion tracking to predict the location and implement data association respectively. Spectral features are extracted for each vehicle to solve the limitation of motion tracking through feature matching. The Bhattacharyya distance is used as a criterion in the feature matching procedure. Our algorithm has been tested using three sets data. One is a set of multispectral polarimetric imagery acquired by the Multispectral Aerial Passive Polarimeter System MAPPS). Another two data sets are spectropolarimetric imagery generated by the Digital Imaging and Remote Sensing Image Generation tool. The tracking performance is analyzed by calculating performance metrics: track purity and Multiple Object Tracking Accuracy ) MOTA. For MAPPS data,the average MOTA and track purity of feature-aided tracking increase 1 percent and 9 percent over those of motion-only tracking respectively. For DIRSIG data with trees,the average track purity of feature-aided tracking in without noise case increases 2 percent over that of motion-only tracking. In this work,we have demonstrated the capability of detection and tracking methods applied in a complex environment.