Generating Maps from Aerial and Street View Imaging

Use case

The client, a mapping company, sought to automate the labor-intensive process of manually identifying vertical traffic signs in panoramic street view images. This manual process, which required browsing all images from the captured route, identifying objects, clicking on them, determining their position in the point cloud, and writing attributes into a database, was a non-trivial, lengthy, and time-consuming bottleneck in their map generation workflow.

Solution

We designed a computer vision system to detect vertical traffic signs within panoramic images. The system leverages a machine learning model trained on a dataset of annotated street view images. The model accurately identifies the presence of traffic signs, reducing the need for human intervention.

Result

The designed system would reduce the manual workload associated with traffic sign identification. Also, the system is planned to efficiently process large volumes of street view imagery, enabling the client to generate maps more rapidly and cost-effectively.

With the core of the system in place, the client will also be able to gradually increase the scope of the system – so that it detects most or all of the 50+ other geospatial object types they annotate and show on their maps.

The AI system also unlocks the possibility of cost effective regular updates of the existing maps. As the system can highlight the changes compared to the previous version, this dramatically reduces the associated service costs for the client.