The vast majority of the annotated datasets used in research and academia lacks any labeled radar data. Some collections include automotive radar information but those are usually represented on a high-level by sparse point clouds, having lost valuable information by preprocessing steps. This application, written in modern C++, OpenGL and Qt marks the attempt to classify low-level radar data in a similar fashion to Lidar and Camera images. After semi-automatic projection of the Lidar points into the camera image and assignment of the corresponding semantic classes, those points vertical projection if used to assign unique class labels to each range-azimuth cell of the radar cone. This work is by no means finished, as it turned out that annotating radar frequency plots is far more involved than initially anticipated. Also, the field of self-supervised learning seems to be a more natural fire for domains in which labeled data are scarce or just hard to come by. Nonetheless, the project contains many valid assumptions and showcases a systematic approach how to obtain annotated radar plots for the purpose of using them in the field of supevised learning.

Labeltool used to annotate real-world automotive radar frequency information