AUTOMATED INFRASTRUCTURE DAMAGE DETECTION

The traditional way of maintaining our roads and other infrastructure is based on manual assessments or inspections which have been subjective, error-prone (due to human factors), and require considerable effort. This proposal will showcase CrackAlert as an automated system that will revolutionise the monitoring of infrastructure using computer vision and deep learning technology. Through a high performance software pipeline, CrackAlert converts the raw images of roads into engineering data with real-time identification of structural failures (eg. longitudinal cracking, potholes, alligator cracking). The implementation of the CrackAlert system is focused on a very strong web-based portal that connects the complex artificial intelligence (AI) models with practical field applications. Each of the components of the software utilize a dedicated HTML5 based canvas rendering engine, with pixel-wise semantic segmentation used to precisely identify damage. This visual data is combined in real-time and geolocalised (using the Web Geolocation API) with GPS data, so that each identified defect can be mapped and classified by severity immediately. The CrackAlert platform is hardware agnostic and scalable, allowing civil engineers to ‘shift from reactive maintenance to data driven proactive maintenance’. CrackAlert provides a low cost method for improving urban safety and preserving modern transportation networks.