A Visual Relocation Algorithm With Illumination Changing Robustness

The visual slam algorithm is mainly used for estimating the state of the camera and building a map for the scene, including tracking, mapping, loop closure detection and relocation. The relocation algorithm is to obtain the position and posture of the current camera in the constructed global map. One of the disadvantages of visual slam is that it is susceptible to lighting changes of the scene. Severe lighting changes in the scene can cause locating failure easily. In most practical work, the lighting conditions of the current momentary environment are inconsistent with the lighting conditions at the time of map construction, which may lead to relocation failure when we use visual SLAM. Therefore, a visual relocation algorithm is proposed to improve the success rate of relocation under different lighting conditions in this paper.