Abstract: When the ambient light of the robot changes, the performance of the loop closure detection algorithm based on the traditional visual word bag will decrease, and it is prone to perceptual aliasing and perceptual variation, thus judging the false closed-loop. In this paper, the original color image is used to generate an illumination invariant image related only to the light source, and then a visual dictionary of the original illumination invariant image is generated. For each image, two histograms and similarity scores are calculated to determine whether it is a closed loop. Finally, it is tested on the data set. The experimental results show that compared with the bag-of-words (BoW), the loop closure detection algorithm proposed in this paper has better robustness to the changes in the environment.