Monocular Visual Odometry Method for Indoor Mobile Robots

With the recent advancements in artificial intelligence, various intelligent robots, including floor cleaning robots, surgical robots, and industrial robots, have become increasingly prevalent in our daily work and lives. These robots require not only a high level of intelligence but also a keen perception of their surroundings. This research is dedicated to the design of a monocular visual odometer system tailored to the navigation and positioning requirements of low-cost indoor mobile robots. The monocular vision odometer offers a simple yet cost-effective and real-time solution, enabling stable operation of mobile robots within indoor environments. To address the task of monocular visual image depth estimation, we introduce a novel monocular visual odometry system in this research. In our system, we propose an ORB algorithm that integrates scale information. Moreover, to overcome the issue of feature points’ main direction instability in traditional ORB algorithms, we incorporate grayscale information surrounding the feature points and employ a new window model to determine the main direction, thus enhancing stability. In the feature matching phase, we used the random sampling consistency (RANSAC) algorithm and make improvements to reduce its computational time. Finally, we conduct experiments to validate the effectiveness of the proposed algorithm. The experimental outcomes demonstrate that our method significantly enhances camera pose recovery accuracy while maintaining real-time performance for the mobile robot system.