在线判别式超像素跟踪算法

针对传统超像素跟踪方法中建模速度慢、目标遮挡易漂移的问题，提出一种新的目标跟踪算法．该算法利用超像素分割，获得大量目标前景和背景的超像素训练数据，通过训练超限学习机，并结合k-d树聚类快速构建目标前景和背景的判别式模型．在跟踪过程中，利用构建的模型和粒子滤波估计目标中心位置．最后结合相关滤波估计目标尺度，实现对目标的鲁棒性跟踪．实验结果表明，所提算法具有可靠的跟踪精确度和较快的跟踪速度．

Online discriminant based superpixel tracking method

In order to solve the problems of slow modeling speed and target occlusion easy drift of the traditional superpixel tracking method, a new target tracking method is proposed. The new proposed method uses superpixel segmentation to obtain a large number of target foreground and backgroundsuperpixel training data. By training the extreme learning machine and combining with the k-d tree clustering we can obtain a discriminant model of the target and background discriminant model quickly.In the tracking process, the constructed modeland the particle filter are used to estimate the target center position. Finally, the target scale information is estimated by correlation filtering to achieve the robustness tracking of the target. Experimental results show that the proposed algorithm has a reliable tracking accuracy and a fast tracking speed.