Visual Loop Closure Detection over Illumination Change

In the Simultaneous Localization and Mapping (SLAM) problem, loop closure detection is a task of whether the robot has visited the area or not, when robot has traveled a long distance, and then revisits the previous travel route. Bag-of-visual-words method, one of the popular and fast visual loop closure detection method, converts a query image into a descriptor and compares it with the descriptors of the whole database images to determine loop closure. However, bag-of-visual words method has lower performance when the illumination is changed while a long driving time because the characteristic of the image is changed due to the illumination difference. In this paper, we propose a novel loop closure detection method robust to illumination change through fusing the results of parallel loop closing detection in original color space and illumination invariant space both by generating illumination invariant space image based codebook. Experimental results show that the proposed algorithm robustly performs loop closure detection over illumination change than conventional method.