Скачать или смотреть RTCnet: A Radar based, multi-class, single-frame road user detection method

We present a novel radar based, single-frame, multi-object, multi-class detection method for moving road users (pedestrian, cyclist, car).
Legend: green/red/blue means pedestrian/cyclist/car predicted classes.

On this video, we show example scenes from our testing set with object-level results overlaid. Classified radar targets are shown as colored circles mapped to the camera image at the radar's height. Segmented objects (i.e. a group of targets belonging to the same instance) are visualized by a convex hull around them. The hull is both plotted on the ground and at 2m height to give 3D appearance.

Camera is for visualization only, every detection shown is strictly based on (single-frame) radar data. No temporal filtering or tracking is applied.