Municipal Infrastructure Anomaly and Defect Detection

Road quality assessment is a key task in a city's duties as it allows a city to operate more efficiently. This assessment means a city's budget can be allocated appropriately to make sure the city makes the most of its usually limited budget. However, this assessment still relies largely on manual annotation to generate the Overall Condition Index (OCI) of a pavement stretch. Manual surveying can be inaccurate, while on the other side of the spectrum a large portion of automatic surveying techniques rely on expensive equipment (such as laser line scanners). To solve this problem, we propose an automated infrastructure assessment method that relies on street view images for its input and uses a spectrum of computer vision and pattern recognition methods to generate its assessments. We first segment the pavement surface in the natural image. After this, we operate under the assumption that only the road pavement remains, and utilize a sliding window approach using Fisher Vector encoding to detect the defects in that pavement; with labelled data, we would also be able to classify the defect type (longitudinal crack, transverse crack, alligator crack, pothole ... etc.) at this stage. A weighed contour map within these distressed regions can be used to identify exact crack and defect locations. Combining this information allows us to determine severities and locations of individual defects in the image. We use a manually annotated dataset of Google Street View images in Hamilton, Ontario, Canada. We show promising results, achieving a 93% F1-measure on crack region detection from perspective images.