基于外观和深度信息的视觉跟踪算法研究

Visual tracking algorithm based on appearance feature and depth information

能够实现视觉导航的自主移动机器人具有很好的应用前景,而场景变化、目标运动、障碍、遮档等是自主机器人视觉导航过程经常遇到的问题,结合外观特征和深度信息的目标检测和跟踪算法是提高自主机器人对目标及环境变化适应能力的重要途径。文章结合人类在跟踪和定位目标时既利用颜色、亮度、形状、纹理等外观特征,又利用物体间距离、深度信息的特点,提出了结合外观特征和深度信息的目标跟踪算法并通过实验验证了该算法对视角、运动、遮挡等因素所引起变化的适应能力,且利用定量的方法对算法的性能进行了评价。

Challenges that robot faces in vision-based navigation include scene change, appearance change, obstacle, occlusion etc. Imitating human vision perception, an object detection and tracking algorithm that combines appearance feature and depth information is proposed. First, RGB image and depth information are captured by the Kinect camera that works as the vision system of robot. Then, an appearance model is created with features extracted from RGB image. A motion model is created on plan-view map produced from depth information and camera parameters, and the estimation of object position and scale is performed on the motion model. Finally, appearance features are combined with position and scale information to track the target. Experimental result show the robustness of our object detection and tracking method to appearance changes arose from view, motion and occlusion factors. It also shows that the object detection efficiency and object tracking accuracy are improved greatly compared with the method that only employ the appearance features.