Robustness Improvement of Long Range Landmark Tracking for Mobile Robots

In this paper a landmark tracking system for mobile robot navigation is introduced focusing on the aspects of stability and robustness. The tracking system uses template matching on selected landmarks complemented with visual odometry based motion estimation. The disparity map created by stereo matching togetger with a saliency map is used to evaluate the areas around the robot based on their suitability for landmark selection. These landmarks are tracked using ZNCC (Zero-Mean Normalized Cross-Correlation) template matching aided with UKF (Unscented Kalman Filter) to estimate and narrow down the search area. The condition of each landmark is evaluated in every frame and the system automatically omits current landmarks and extracts additional landmarks if needed. The criteria for this is based on the template matching results and the travel trajectory of the robot.