"The Pedestrian next to the Lamppost" Adaptive Object Graphs for Better Instantaneous Mapping
This work addresses a key bottleneck in autonomous navigation by improving BEV mapping accuracy, particularly for distant objects, though it is incremental as it builds on existing texture-based models.
The paper tackles the problem of increasing localization error with distance in bird's-eye-view (BEV) mapping from monocular images by learning spatial relationships between objects, achieving a 50% relative improvement on nuScenes and setting new state-of-the-art results across three datasets.
Estimating a semantically segmented bird's-eye-view (BEV) map from a single image has become a popular technique for autonomous control and navigation. However, they show an increase in localization error with distance from the camera. While such an increase in error is entirely expected - localization is harder at distance - much of the drop in performance can be attributed to the cues used by current texture-based models, in particular, they make heavy use of object-ground intersections (such as shadows), which become increasingly sparse and uncertain for distant objects. In this work, we address these shortcomings in BEV-mapping by learning the spatial relationship between objects in a scene. We propose a graph neural network which predicts BEV objects from a monocular image by spatially reasoning about an object within the context of other objects. Our approach sets a new state-of-the-art in BEV estimation from monocular images across three large-scale datasets, including a 50% relative improvement for objects on nuScenes.