For many years, the City of Cape Town had undertaken manual surveys of their road segments, a long and arduous task to inform which sections required maintenance work. This would require surveyors conducting on-site inspections and surface area calculations of roads targeted for repair or resurfacing.
Aurecon was appointed to help the City of Cape Town with maintenance of their road networks’ budgeting and planning by surveying the road widths for all metropolitan roads.
While a myriad of legacy information about Cape Town's metropolitan road network already existed, the data was not sufficiently reliable or consistent and so manual surveying remained standard practice.
With the total length of the city’s road network estimated to be 8 200 kilometres with 101 152 road segments, updating data of the entire network by manually measuring each road segment would have been high risk, extremely time consuming and financially prohibitive.
New advancements in technology provided the ability to digitally capture and determine these measurements and provided an opportunity for Aurecon to develop a new approach.
By combining new technology with our surveying, geographic information system (GIS) and machine learning expertise, a new workflow was established that included the creation of a machine learning model that extracted road surfaces from high resolution aerial images. With Aurecon's experts located in three different locations, a team was set up across Cape Town, Melbourne and Auckland offices, to develop and test an approach that captured and validated new and existing measurements of the metropolitan road network.
The images taken of the roads became building blocks used to train and test the machine learning model to classify each pixel as either road or non-road. This helped determine road surface areas per road segment, and the accuracy of each segment.
The scope encompassed a range of road categories, each with varying quality and visual detail from various sources including aerial photography, laser scanning and geospatial metadata.
The machine learning model achieved the desired accuracy for the majority of the road network. This new workflow delivered the road surface area calculations in shorter timeframes compared to a manual approach with the necessary levels of confidence and accuracy, enabling the engineers to focus on constructing a well-connected city.