Visual Loop Closure Detection with Scene Mutual Information for Mobile Robot

In this paper, an efficient approach is proposed for loop-closure detection in robot visual SLAM. The method uses mutual information to measure similarity between current view and key frames in an appearance map, and evaluates candidate loop-closure locations in particle filter framework. Specially, the implementation of particle filter is accelerated through updating a set of weight vector of particles, and three threshold indicators are used to select loop-closure candidates and verify loop-closure location. The comparative experiments on a popular dataset verify the high efficiency of our method which is more simple and accurate than the popular bag-of-words (BoW) for loopclosure detection.