Crack Detection Based on Support Vector Data Description

It is difficult to detect concrete cracks because of the existence of background interference. To solve this problem, some methods for crack detection on concrete surfaces are analyzed. According to shape features, a new concrete crack detection method is proposed. First of all, iteration method is applied to get the optimal threshold for image segmentation after grayscale transformation. Secondly, binary images are processed by morphological closing operation and deburring. Features including eccentricity, circularity and packing density are selected as input training vectors for Support Vector Data Description (SVDD). The experimental results show that crack detection method based on SVDD can accurately distinguish cracks from other kinds of defects (non-crack) and reduce the false negative detections.