Markovian-based detection and extraction of defects on textured surfaces by spatio-temporal video processing

This paper proposes a method for the detection and extraction of poorly contrasted linear structures in textured flat areas. The method is a spatio-temporal processing based on an irregular Markov random field (MRF) modeling of the image plane. It combines in an appropriate way a pure intra-frame processing based on an irregular graph-based MRF modeling and a time recursive nonlinear filtering which provides the final topological structure of the defects. The application considered here is the detection and classification of various defects that may appear on a roadway, particularly cracks. The images are produced by a video camera on board a running vehicle and showing a perspective view of the pavement. The results produced by the spatio-temporal processing are significantly better than a purely spatial analysis method gives, as taking into account multiple views of the same section of a roadway allows one to increase a lot the ability to discriminate between distressed and fair textured pavements