Seite - 780 - in Book of Full Papers - Symposium Hydro Engineering

past surveys. The results of the automatic defect identification is hence dependent on the scope of effort spent on labeling the data (in terms of spatial or semantic information). 3.2. IMAGE-PROCESSING FOR DEFECT DETECTION In order to formulate an algorithm, which is capable to detect damages on the concrete surface based on OVC photos, knowledge about the visual appearance of defects is necessary. For example, white spots on a concrete surface usually imply sinter formation, which may result from cracking or block joint openings. Wet areas, pop outs and other deficiencies evoke recognizable visual irregularities. Consequently, we analyze the captured OVC photos on these characteristics. While most researchers focused on detecting cracks from images [8, 9, and 10] in the past, we generalize our analysis to detect any kind of concrete deficiencies on a dam using the captured OVC images. Fig. 7 shows the intermediate results of the algorithm. Detailed information on the procedure can be found in [11]. The algorithm describes potential defects in terms of pixel coordinates of the bounding rectangle (Fig. 7d), which eventually gives the orientation angles for the TS (using the known camera model). This angle values are used to steer the TS and to capture high-resolution images of the potential defect areas. It becomes obvious that the complete data acquisition is automatable by using a modern TS and a database with spatial information on the dam’s boundaries and apparent defects. (a) Original OVC image (b) Segmented gradient image (c) Grouped image regions (d) Filtered and identified defects Fig. 7: Important steps of the defect detection algorithm 780