FAST Algorithm for Accurate Corner Points Detection of Section Steel Based on Adaptive Threshold

Abstract

The on-line flatness detection of section steel based on machine vision is a key technical problem for quickly and accurately extracting key corner points from section steel images to enable accurate detection. Aiming at the problem that the features from accelerated segment test (FAST) algorithm needs to manually set the corner points screening threshold and there are numerous false corner points in corner point extraction, this paper proposes an adaptive threshold generation and correction strategy. Based on the automatic determination of the initial threshold, this strategy can adjust the threshold in real time until an appropriate value is reached according to the requirements of the initial corner points set, thereby to reduce the risk of missing key corner points. In addition to using FAST algorithm to extract corner points, the smallest univalue segment assimilating nucleus (SUSAN) algorithm is employed to eliminate false corner points ensuring the effectiveness of key corner points extraction. The experiments prove that the FAST corner detection algorithm based on adaptive threshold (FAST-A) can still accurately and quickly detect key corner points even when the detection environment and object characteristics change. Furthermore, the algorithm proposed provides accurate corner points for real-time section steel flatness detection, and improves the adaptability of corner points extraction.

Keywords

• section steel
• corner points detection
• features from accelerated segment test (fast) algorithm
• smallest univalue segment assimilating nucleus (susan) algorithm
• adaptive threshold