1.0 Introduction and Background
The maintenance and management of our degrading and continuously ageing built infrastructure is a majorissue. Structural Health Monitoring (SHM) can be a sustainable and efficient solution in this regard. Imagebased techniques offer a relatively inexpensive approach to SHM and often a standard digital camera may bethe only major capital investment. An additional advantage of this ideal non-contact method is that theimages captured of a particular structure or damaged region may be archived for later reference.The introduction of an imaging protocol can help alleviate confusion which may arise on the part of thephotographer who may often not be an expert related to the object of interest. An example is the applicationpresented in this thesis where a diver carries out underwater inspection and imaging of corroding steel piles.Most often, the diver is not an expert or even an engineer. The diver’s only fulfilment is to record videofootage of the corroding piles as detailed by the engineers guidelines, there is no requirement on his/her partto interpret the captured data.Although the human eye is capable of detecting a much wider range of tone and lighting, human errors andsubjectivity often undermine it, whereas an image based method may be far more reproducible, objectiveand conducive to post-processing. Even with successful detections through image processing, there alwaysremain uncertainties related to Probability of Detection (PoD) and Probability of False Alarm (PFA) withany NDT technique and these two quantities establish the efficiency of the technique. The plot of PoDversus PFA is called a Receiver Operating Characteristic (ROC) curve and this curve establishes theperformance of an NDT. The best performance that can be achieved by any method indicates a certainprobability of detection (1.0) and no false alarm (0.0). The distance of any ROC for a particular NDTtechnique from this point with PoD and PFA as the axes is the Best Performance Point (BPP). This distancealso provides the efficiency of performance from an absolute and a relative point of view.A key challenge associated with image based detection is related to image stabilisation, a challenge which isexacerbated in underwater conditions as a stable anchorage point for the camera is generally unavailable.Additionally, poor environmental conditions, obstructions in the field of view or access problems mayinhibit the use of a camera. This thesis addresses the problem of corrosion detection for steel piles underpoor lighting conditions by proposing HDR imaging in conjunction with traditional and novel featurerecognition techniques. The features of interest should remain static for HDR imaging and this condition isgenerally satisfied when the targets are corrosion regions of static piles when the camera movement is notsignificantly high.