A Novel Tag-Based Robust Pose Estimation and Tracking Algorithm via Binoculars Visual Measurement

Designing a visual feature-based implementation of robot simultaneous localization and mapping (SLAM) is challenging, particularly for achieving real time, high-precision state estimation and tracking. In this article, we propose a novel image label-based alignment algorithm combining one-versus-the-rest support vector machine (OVR-SVM) and peak signal-to-noise ratio (PSNR) similarity to enhance accuracy and robustness in measurements from calibrated binocular vision sensors. Designed to enhance real-time performance in the template-based matching process, a set of novel simple labels is employed. The relative position relationship between these labels, determined during the initial estimation, serves as a position constraint in the back-end optimization process, eliminating outliers and expediting the state estimation and tracking of the robot. Subsequently, an experiment comparing average times is conducted to validate the improved computational rate of the binocular vision system. Finally, the experimental platform of the label-based vision robot is built to operate the motion process of robot to verify the real time, accuracy, and robustness of the proposed pose estimation and tracking algorithm.