A Line Extraction Process Based on Noise Elimination and a Novel Benchmark in Jensen-Shannon Divergence

This paper proposes a line extraction process based on noise elimination and a novel benchmark in Jensen-Shannon divergence. For com-mon noise from sensors, DBSCAN clustering im-proves accuracy and time efficiency in line extraction. Extracted line segments are further adjusted according to our new benchmark based on JS divergence, which is also used to finally evaluate the results against the ground truth. As for the crucial line extraction algorithm, which is a replaceable part, we optimize Split-and-Merge algorithm to split by sum of variance or weight. In comparison with four classic algorithms, three steps in our new pipeline contribute to producing better outcomes in different dimensions.