A line segment detector based on subanchor point and line segment combination method

Abstract

Linear segment features contains rich geometric information and it is crucial to perform accurate linear segment detection. We propose a temporal line segment detector based on subanchor point and line segment combination approach, which gives more accurate results without adjusting the parameters. The speed remains essentially the same as the fastest running line segment detectors (EDLines) existing and is 89’ faster compared to the conventional LSD line extraction algorithm. The proposed algorithm introduces bilateral filtering and line segment combination methods based on EDLines. The bilateral filter is used after the image input to remove the noise and retain the edge information of the image well. The short line segments that exist homologous are combined to generate clearer long line segments, which overcomes the problem of broken lines in LSD algorithm and excessive filtering of short line segments in EDLines algorithm. We validate the effectiveness of the proposed algorithm in experiments of line feature extraction for multiple sets of images in different scenes.