Automatic bridge crack detection - A texture analysis-based approach

Abstract

To date, identifying cracks in bridges and determining bridge conditions primarily involve manual labour. Bridge inspection by human experts has some drawbacks such as the inability to physically examine all parts of the bridge, sole dependency on the expert knowledge of the bridge inspector. Moreover it requires proper training of the human resource and overall it is not cost effective. This article proposes an automatic bridge inspection approach exploiting wavelet-based image features along with Support Vector Machines for automatic detection of cracks in bridge images. A two-stage approach is followed, where in the first stage a decision is made as whether an image should undergo a pre-processing step (depending on image characteristics), and later in the second stage, wavelet features are extracted from the image using a sliding window-based technique. We obtained an overall accuracy of 92.11% while conducting experiments even on noisy and complex bridge images.