RIMM: A novel map matching model with rotational invariance

The Map Matching Problem (MMP) aims to find a real optimal region from a map repository, which is the most similar map to the ideal sample. Though there is a significant difference between MMP and the general Image Matching Problem. The former prefers approximate match and ignores details of edge. Both of them must solve variances of scale, translation and rotation. However, the Scale Invariant Feature Transform (SIFT) is not perfect to match the rotated maps because the number of feature points selected by SIFT is really variant while the image is rotated. This paper presents a simple novel Rotational Invariant Map Matching (RIMM) model that is really invariant to scale, translation and rotation. The RIMM model achieves these invariance abilities based on the inherent, stable, and unique attributes of a map, which are mass center and Primary Gradient Direction. Because these two features are irrelevant to translation, rotation and scale. Our test results show that the RIMM model accurately eliminates variance of translation and rotation. Moreover, The RIMM model can quantitatively evaluate the matched maps by the cosine similarity so that the candidate maps can be ranked and compared objectively.