Advances in visual SLAM research

Visual SLAM, i.e., simultaneous localization and mapping with cameras, plays an important role in the navig-ation of robots, unmanned aerial vehicles, and unmanned vehicles. As the location accuracy affects the obstacle avoid-ance accuracy and the mapping quality directly affects the path planning performance, the visual SLAM algorithm is the core aspect of intelligent mobile applications. This paper introduces the architecture of the mainstream visual SLAM system, including several common visual sensors, the function of the front end, and the optimized back end. According to the type of the metric map model created by the visual SLAM system, visual SLAM can be classified into three types: sparse visual SLAM, semi-dense visual SLAM, and dense visual SLAM. The landmark achievements and research pro-gress of visual SLAM are reviewed in this paper, and its current problems and possible future developments are discussed.