Visual SLAM in Dynamic Environments using Multi-lens Omnidirectional Camera

Many studies relate to Visual SLAM (Simultaneous Localization and Mapping) systems have been conducted for automatic driving and assisted driving systems. In this paper, we propose the Visual SLAM system combining multi-lens omnidirectional camera and object recognition. The proposed method can reduce some noises caused by moving objects, such as vehicle, pedestrian, bicycle, and so on, in the road environment. We also propose a robust SLAM system against tracking loss by improving the tracking thread. Since a multi-lens omnidirectional camera system can view not only the front but also the left and right sides of the scene, it is possible to obtain effective feature points even in scenes with little texture. Therefore, it provides better SLAM results than monocular camera systems.