Affine-Invariant Geometric Constraints-Based High Accuracy Simultaneous Localization and Mapping

Figure 1

Examples of two problems occur in [4, 5, 7, 8] on the basis of CityCenter [14] dataset. (1) illustrates that, because too many words are matched between two different locations incorrectly, previous pure bag-of-words systems treat two different locations as the same place. By using the proposed method, 0 match is accepted. So the proposed method can solve this problem. (2) shows Raw-Likelihood (without normalizing) compression between [5] and the proposed method. Although the ground truth obtains the highest likelihood, because there are incorrect matches between two many different locations, those noises result in that the ground truth is rejected by previous pure bag-of-words systems incorrectly. The proposed method prevents noises and retains the peak of ground truth and thus the proposed method can accept the ground truth correctly.