Longest similar subsequence-modified Hausdorff distance: A new robust approach for measuring distances between linear features

Abstract

Accurately measuring the distance between linear features is crucial for understanding the similarity between objects. Traditional measures such as the Hausdorff distance (HD) are inadequate in handling length variations and outliers among linear features, and thus reducing the accuracy of similarity assessment. In this paper, we propose a new concept termed Longest Similar Subsequence (LSS), which aims to identify the longest overlapping subsequence of paired linear features. By integrating LSS with a modified Hausdorff Distance (LSS-HD), we propose a matching method that is based on LSS-HD and adaptive threshold-selection. The adaptive threshold-selection method, named the Connected Bin Threshold Selection Algorithm (CBTSA), is based on the distance histogram analysis. Experimental results clearly demonstrate that the LSS-based HD measure is more robust compared to the HD measure without LSS constraints. In the road matching experiments, our matching accuracy, with an F1-score of over 0.96, is significantly higher than that of the matching algorithms (Opt and OILRM), which are also solely based on distance-similarity metrics.