MASSIVELY-PARALLEL COMPUTING TECHNIQUES MIGHT IMPROVE HIGHWAY MAINTENANCE

Manual methods of measuring defects in roads show poor repeatability a nd reproducibility. Cracking is a principle indicator of defect progre ssion in a road pavements, and the authors' overall objective is to de velop a practical automatic, repeatable, and reproducible method of de termining the extent of cracking. Their research aims at using a distr ibuted array of processors to achieve practical speeds for processing digitized images of road surfaces to detect cracks. The algorithms des cribed here provide for two processes. The first converts a gray-scale image into a binary image that represents most of the cracks and elim inates most of the noise from the surface texture. This initial screen ing process might suffice for the bulk of a road having few cracks. Th e second process combines the crack fragments in the binary image into continous cracks and gives the highway engineer an appropriate output . The article includes results in which individual images were judged to contain cracks or not contain cracks by eight independent observers and by processing on the DAP to the end of the initial screening proc ess. The authors have found that single images can be processed to the initial screening stage in the 40-millisecond limit for real-time pro cessing provided by the British TV standard.