Detection of Curves with Unknown Endpoints using Minimal Path Techniques

We present a novel method to detect curves with unknown endpoints using minimal path techniques. Our work builds on the state of the art minimal path techniques currently used to detect curves. Existing algorithms in the literature require the user to specify both endpoints of the curve or one endpoint plus the total length of the curve. However, in our approach, the user may specify any arbitrary initial point on the curve and the algorithm can detect the complete curve (even with multiple branches) automatically without the need for any additional information. We apply this algorithm to the problem of crack detection in civil structures where cracks are modeled as 2D curves. The results demonstrate that the algorithm is robust to variations in background and texture and is able to detect curves accurately. ------------------------------------------------

• Type: Journal
• Year: 2010
• Where: BMVC
• Authors: Vivek Kaul, Yichang Tsai, Anthony J. Yezzi
• Keywords: Algorithm, BMVC 2010, Computer Vision, Curve, Minimal Path