Vision-based Moving Target Detection and Tracking Using a Quadrotor UAV

Autonomous detecting and tracking of mobile ground robots are two fundamental tasks for the cooperation between quadrotor UAVs and UGVs. In our work, a T-shape landmark is installed on the top of a ground robot so that the quadrotor can use onboard monocular camera to perform fast object detection. Considering the effects of shading, shadows and varying lighting condition, the adaptive thresholding algorithm and topology analysis on a binary image are adopted to accomplish online T-shape landmark detection. Since both the quadrotor and the ground robot are in the continuous moving state, a hybrid particle filter algorithm is presented to implement robust object tracking with a low computational cost. To solve the problem of particle measurement matching optimization, a method of achieving least uncertainty measurement based on Mahalanobis rule and minimum Euclidian distance is utilized in this paper. A series of experiment results with monocular vision in an indoor environment show our approach's validity.