A Monocular SLAM System Based on the ORB Features

Abstract

With the rise of the AR/VR industry and the development of intelligent mobile robots and driverless vehicles, the industry has seen a spurt in demand for SLAM technology. On the other hand, the development of sensor technology, the abundance of computing resources, and the advancement of algorithms make SLAM technology transition from algorithm to product. The visual SLAM technology relies on the small size, low power consumption, and information-rich visual sensors to provide a map of the environment. Its positioning results in the map for robots in unknown environments, which is of great significance for robot automation and intelligent applications. Among the many vision SLAM technologies, ORB-SLAM stands out for its outstanding performance as an open-source project and a complete SLAM system that incorporates various current mainstream approaches. This paper presents the main components and methods of Monocular ORB-SLAM and tests and evaluates them using current popular datasets. Monocular ORB-SLAM can perform well in slow-movement, textured scenes but still needs to overcome the issue of tracking lost due to camera shake and fewer textures.