Development of UAS-based Construction Stormwater Inspections & Soil Loss Model
Stormwater management practices on construction sites provide effective solutions for minimizing the environmental hazards that occurs due to sediment runoff associated with construction activities. Construction sites tend to discharge significant amount of sediment due to the nature of land disturbing and grading activities. The implementation of erosion and sediment control practices is required by federal, state, and local regulations. These regulations require regular inspections of the practices for evaluating their performance. Inspection requirements consist of qualified personnel, frequent inspections, corrective actions, and record keeping. This study integrated innovative tools and guidance into aerial stormwater inspection procedures for improving current stormwater inspection and design approaches. Trial inspections were conducted at the U.S. highway 30 construction site in Tama County, Iowa. The use of Unmanned Aerial Systems (UASs) in aerial inspections brought the capability of collecting vast number of geo-located images in a short period of time. These images were processed by using photogrammetry applications to produce orthomosaic view and the digital surface model (DSM) of the area of intent. The processing results enabled data analysis with the integration of geographic information systems (GIS), CAD applications and deep learning-based object detection. The use of GIS assisted data analysis by providing site plans, hydrologic analysis and soil loss model. Orthomosaic views were used as datasets for site plans and object detection, while DSMs were used for hydrologic analysis and soil loss model. Hydrologic analysis provided the opportunity for evaluating the performance of the practices. CAD applications were used to calculate the catchment areas on the surface. Soil loss model production, which utilized RUSLE equation in GIS, showed the potential to support design approaches by determining vulnerable areas for erosion on the site. Moreover, deep learning-based object detection principles were integrated into aerial inspections for automating inspections steps. This approach included training a code with large number of annotated images so that the code can localize and classify practices by using aerial imagery. Deep learning-based object detection is capable of gathering information on the number, location and the type of practices. This feature automates inspections and supports design procedures for the comparison of site conditions with plan sets. This research introduces aerial inspections as an innovative and effective method to improve current inspection methods, which are not sufficient to meet the inspection requirements. Aerial inspections save time and resources by using fewer inspectors, providing better record keeping, having faster inspection procedures and developing efficient data analysis results to evaluate the performance of practices. The study highlights the potential for this technology and developed approaches to be used in the construction industry.