Municipal Road Infrastructure Assessment Using Street View Images

Image Processing;Machine Learning;Image Segmentation;Road Segmentation;Crack Detection;Infrastructure Assessment;Pavement Monitoring

Abou Chacra, David

University of Waterloo

关键词: Image Processing;    Machine Learning;    Image Segmentation;    Road Segmentation;    Crack Detection;    Infrastructure Assessment;    Pavement Monitoring;   

Others  :  https://uwspace.uwaterloo.ca/bitstream/10012/10899/3/AbouChacra_David.pdf

瑞士|英语

来源: UWSPACE Waterloo Institutional Repository

【 摘 要 】

Road quality assessment is a crucial part in Municipalities;; work to maintain their infrastructure, plan upgrades, and manage their budgets. Properly maintaining this infrastructure relies heavily on consistently monitoring its condition and deterioration over time. This can be a challenge, especially in larger towns and cities where there is a lot of city property to keep an eye on. Municipalities rely on surveyors to keep them up to date on the condition of their infrastructure to prevent this failure before it happens. This is both to prevent injuries and further damage from occurring as a result of infrastructure failure, and since it is can be more cost effective to maintain property rather than have to replace it. Surveying can either be done manually or automatically, but it is not done frequently as it is expensive and also time consuming. Manual surveying can be inaccurate, while a large portion of automatic surveying techniques rely on expensive equipment. To solve this problem, we propose an automated infrastructure assessment method that relies on Street View images for its input and uses various computer vision and pattern recognition methods to generate its assessments.First, we segment the image into ;;road;; and ;;background;; regions.We propose a road segmentation algorithm specifically aimed at segmenting roads from street view images. We use Fisher vectors calculated on SIFT descriptors to encode small windows extracted from the main image at multiple scales. Then we classify these patches using an SVM and utilize a Gaussian voting scheme to obtain a segmentation. We additionally utilize a spatial prior to improve this segmentation. Optionally, we improve the segmentation further by making use of a weighted contour map calculated on a shadow-free intrinsic image, and a find an optimal segmentation by utilizing a purity tree. Our algorithm performs well and outputs a good segmentation for further use in road evaluation. We test our method on the KITTI road dataset, and compare it to the state-of-the-art on this dataset, along with a manually annotated subset of Google Street View.After segmenting the road, we describe an algorithm aimed at identifying distressed road regions and pinpointing cracks within them. We predict distressed regions by re-using the computed Fisher vectors and classifying them with a different SVM trained to distinguish between road qualities. We follow this step with a comparison to the weighed contour map within these distressed regions to identify exact crack and defect locations, and use the contour weights to predict the crack severity. Promising results are obtained on our manually annotated dataset, which indicate the viability of using this cost-effective system to perform road quality assessment at a municipal level.