Robust crack detection for unmanned aerial vehicles inspection in an<i>a-contrario</i>decision framework

Aldea, Emanuel; Le Hégarat-Mascle, Sylvie · 2015 · Crossref

DOI: 10.1117/1.jei.24.6.061119

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of robust crack detection in images acquired by unmanned aerial vehicles (UAVs), which are often degraded by motion blur, defocus, and illumination variations. The authors identify a gap in existing literature: while many crack detection algorithms perform well on high-quality data, they rely on heuristic thresholds that fail under the significant image degradation typical of UAV inspections. To solve this, the authors propose a detection strategy based on *a-contrario* modeling, which identifies structures by modeling the absence of features (noise) rather than the features themselves, thereby eliminating the need for manual threshold tuning. The proposed method operates in two main stages: local seed selection and global reconnection, both governed by statistical significance. First, the authors preprocess images using background subtraction and steerable filters to enhance linear structures. They then define a probabilistic measure based on the Number of False Alarms (NFA), which quantifies the likelihood that a detected structure occurred by chance under a naive noise model. This allows for the identification of "atypical" pixels (seeds) without fixed thresholds. Second, the algorithm reconnects these seeds by finding minimal cost paths that maximize their atypicalness relative to the background model. This approach contrasts with traditional methods like Free-Form Anisotropy (FFA) or image percolation, which require thresholding and struggle when degradation blurs the statistical distinction between cracks and background. Experiments were conducted on real image datasets to which complex blur was artificially applied to simulate UAV acquisition conditions. The results demonstrate that the proposed *a-contrario* strategy is significantly more robust than reference methods. While traditional algorithms experienced significant performance degradation and failed to detect cracks in blurred images due to their reliance on sensitive thresholds, the proposed method maintained effective detection capabilities. The study confirms that by avoiding heuristic parameters and relying on statistical contradiction against a noise model, the algorithm can reliably detect cracks even when image quality is severely compromised. The significance of this work lies in its contribution to automated nondestructive examination (NDE) for infrastructure surveillance. By providing a threshold-free, statistically robust detection framework, the method enables scalable, automated inspection systems that can operate effectively in real-world UAV conditions. This reduces the need for human supervision and allows for the reliable monitoring of large-scale infrastructure, such as roads and buildings, despite the inherent image quality limitations of aerial data acquisition.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• signal detection theory