Robust Place Recognition with Combined Image Descriptors

In this paper, a method of place recognition is presented. The method is generally classified under the bag-of-visual-words approach. Information from several global image descriptors is incorporated. The data fusion is performed at the feature level. The efficacy of the combined descriptor is investigated on the dataset recorded from a real robot. To measure the composition effect, all component descriptors are compared along with their combinations. Information on computational complexity of the method is also detailed, although the algorithms used did not undergo a big amount of optimization. The combined descriptor exhibits greater discriminative power, at the cost of increased computational time.