Related Works
These are a few project from other people, with a similar goal as mine.
In 2019 David Yu and Hayley Miller took a look at how to automaticly detect and categorise road features.
They used existing videos and images from Mobile Mapping Systems[MMS], like Google Street View or Open Street Map, to detect signs on roads.
This could be a cost-effective way to do asset-management or damage-assessment.
The main difference is of course the lack of real-time operation. But the methods used to enhance the sign detection could also be applied in a project like mine.
https://medium.com/geoai/road-feature-detection-geotagging-600ea03f9a8
The researchers Vladimir A. Krylov, Eamonn Kenny and Rozenn Dahyot of the Trinity College in Dublin had a similar project.
They put a large focus on calculating the exact position of a detected object, by using images from different angles. Their triangulation method achieved higher precision than the usual method.
This is again not in real time, but nonetheless potentially usefull for real-time-operation.
The researchers from Wuhan University tried to mitigate the problems that arise when doing object-detection from afar.
Most projects that I came across used the YOLOv3-model and used pyhton as their language