Research On The Asphalt Pavement Crack Automatic Detection Technology Based On Multi-scale Analysis

The asphalt pavement crack automatic detection is one of the most challenging research subjects in the domain of transportation information engineering and the pattern recognition. The paper, taking the multi-scale analysis theory as a core, studies the asphalt pavement crack automatic detection from two aspects:multi-scale mathematic morphology and the multi-scale geometric analysis.Firstly, in view of characteristics of the pavement surface image with complicated texture and many noise, the paper implemented the multi-resolution of image segmentation to the primitive asphalt pavement crack image. Also, based on the variance analysis, the paper proposed one kind of local threshold algorithm which aims at eliminating the influence created by uneven illumination in the original image and highlighting the crack region.Then, through analyzing the characteristics of crack image and choosing a suited scale, structure element and morphology operation combination, the paper designed an multi-scale mathematics morphology operator, which can efficiently suppress the noise while enhancing crack region. The paper also detected the edge of enhanced pavement surface gray image, and then removed the noise and the false crack by morphology filter. It divided asphalt pavement image into crack class and non-crack class based on Support Vector Machine, whose classifying accuracy rate has reached to 93.1661%.Finally, the paper applied Beamlet transformation which belong to Multi-scale geometric analysis tool to the asphalt pavement crack automatic detection, and implemented non-structure algorithm of Beamlet transformation. The experimental results show that Beamlet transformation can accurately approach to testing goals from different directions, different positions and different sizes. Therefore, it is an effective detection algorithm for asphalt pavement crack, and has a good anti-noise performance.The achievements of this paper provide a set of new algorithms for the asphalt pavement crack automatic detection, which are able to raise the detection accuracy and have some theoretical significance as well as practical value.