基于连续自适应均值漂移和立体视觉的无人机目标跟踪方法

摘要

为设计一种可自动跟踪、避障的无人机视觉导航技术,将无人机同时搭载双目和单目摄像头:单目摄像头采集无人机相对于被跟踪物体的图像,并采用经Kalman预测器优化的连续自适应均值漂移算法对目标进行有效跟踪;双目摄像头实时采集无人机前进方向上的图像信息,并利用SGM算法计算深度图以分割出无人机前进方向上的障碍物信息.无人机在跟踪目标物体的同时,可自主避开行进方向上的障碍物.实验结果表明,该方法可以有效引导无人机持续、精确地对目标物体进行跟踪,并在跟踪过程中及时躲避前进方向上的障碍物.%The purpose of this paper is to design an approach which lets the UAV automatically track the moving objects and avoid obstacles. In this approach, both the monocular and binocular cameras are installed on UAV. The images of a tracked object are captured by the monocular camera while the continuously adaptive mean-shift (Camshift) algorithm optimized by Kalman predictor is used to continuously track the target. At the same time, the image information in the forward direction of UAV is captured by the binocular cameras, SGM algorithm is used to generate depth maps, which are used to segment obstacles in front of UAV. By this approach, UAV can continuously track a moving vehicle and automatically avoid obstacles on the way of flight in real time. Experimental results demonstrate that this approach can effectively and precisely guide the UAV to track a moving object and avoid obstacles in the route of flight.