Research: Road feature detection

Challenge:

Software Supreme wanted to be able to test various Advanced Driving Assistance Systems ideas. ADAS is a tangible benefit to safety, it requires image processing and benefits immensely from a high performing software that uses the available hardware efficiently.

Solution:

We decided to set up a framework that reads image data from multiple cameras or prerecorded sources and enables us to execute various algorithms on the image data. That way we could iterate quickly and test solutions. The application could run both on an edge device(nVidia Jetson AGX) or a developer workstation.

Result:

With the framework we implemented multiple algorithms:

lane detection based on pure image processing
a couple of lane detection algorithms that use neural networks
single shot detector for object detection

Multiple properties are exposed to the user and multiple views from different stages of the image processing could be displayed so it is easier to tweak the algorithms and get a clearer understanding of how they work.