The project was a research and development project for a major automotive supplier. The task was to evaluate state-of-the-art approaches for 3D object detection and tracking in an automotive context.

In the project, I adapted several state-of-the-art CNNs for object (vehicles, pedestrians) detection to the in-house dataset of the customer, trained them, partly on synthetic data, and evaluated the viability of the methods to their sensors.

I further developed an object tracking, allowing for the tracking of objects over an extended period of time.The project allowed for fusion of a forward-facing camera with a radar, which enabled us to also focus on performance of the system in adverse conditions, such as rain or poor lighting conditions.

In a second project with the same customer, the fusion was done with a lidar sensor. This gave opportunities for much higher accuracy, as a lot of CNNs for object detection in automotive benefit strongly from additional lidar input.