关键词: 道路工程, 识别算法, 图像处理, 路面裂缝, 并行框架, 三维虚拟路面, 立体投影, 裂缝种子

Abstract: In order to detect cracks accurately, precisely and completely with the consideration of stereoscopic features of pavement, an automatic parallel crack recognition algorithm based on 3D virtual pavement was proposed. Firstly, 3D virtual pavement was established by hiding processing and illumination model processing on 1mm/pixel pavement depth images, and four projection images, Ω1~Ω4 , were generated via stereoscopic projection at four angles over it. Then, dimension reduction, crack recognition (intensity verification, symmetry check) and crack connection were conducted on Ω1~Ω4 respectively. Correspondingly, shadowed crack images Ωs1~Ωs4 and reflection crack images Ωr1~Ωr4 were obtained. At last, the crack information of Ωs1~Ωs4 and Ωr1~Ωr4 were fused, examined by depth verification and filtering denoised, yielding the final crack image. Test results based on 255 images (4 096×2 048) show that, the proposed algorithm achieves relatively high precision (averaging 80.34%) and recall rate (averaging 83.89%). It outperforms ADA3D algorithm with an F score of 80.47%. Furthermore, the parallel framework of the proposed algorithm helps parallel programming, which can improve computation speed effectively.

Key words: highway engineering, recognition algorithm, image processing, pavement cracking, parallel framework, 3D virtual pavement, stereoscopic projection, crack seeds