Detecting Road Obstacles by Erasing Them
This work provides a method for autonomous vehicles to detect novel road obstacles, improving safety in unpredictable driving environments.
The paper addresses the challenge of detecting unknown road obstacles without prior training data for each obstacle type. It achieves this by inpainting selected image patches with surrounding road texture to remove potential obstacles, then uses a neural network to identify discrepancies between the original and inpainted patches, indicating an erased obstacle.
Vehicles can encounter a myriad of obstacles on the road, and it is impossible to record them all beforehand to train a detector. Instead, we select image patches and inpaint them with the surrounding road texture, which tends to remove obstacles from those patches. We then use a network trained to recognize discrepancies between the original patch and the inpainted one, which signals an erased obstacle.