分块多特征目标描述子的移动机器人目标跟踪

为解决机器人目标跟踪过程中的遮挡和外观改变等问题, 提出一种分块多特征描述子的方法. 该方法将候选样本分块, 提取图像片的深度、颜色、纹理特征来表示目标构造检测器. 结合目标与机器人的运动构造运动卡尔曼滤波器(MEKF) 作为跟踪器. 跟踪过程中根据目标深度信息调整其尺寸, 结合深度特征及图像片外观相似度进行检测并处理遮挡. 实验结果表明, 该算法对目标的尺度变化、光照改变和遮挡现象具有较强的鲁棒性.

Person tracking for robot using patches-based-multi-cues representation

To deal with the problem of occlusion and appearance changes in person tracking with a mobile robot, an algorithm based on patches-based-multi-cues representation is proposed. The algorithm segments the candidate sample and extracts the depth information, color histograms, texture histograms from each image slice for constructing the detector. A motion extended Kalman filter(MEKF) is obtained by considering the motion of the robot and target. As tracking evolves, the target’s size is adaptively adjusted according to the depth histogram. Furthermore, occlusion is identified by simultaneously detecting the depth features and appearance model. Experiments results show that the proposed approach has better robustness for dealing with the problem of scale changes, illumination variations, and occlusion.