Summary
It can be hard for robots to know when a person is partly hidden (occluded) in camera images, especially since labeling such occlusion by hand is costly and rare. This work shows that by comparing two simple distance measures a camera system already produces, it is possible to tell if a person is occluded without any extra training or labels. The key idea is that the box used to detect a person keeps their full height even if partly covered, but the estimated depth inside that box shifts toward anything in front, revealing occlusion. This method works well on standard driving datasets and synthetic scenes and is not fooled by just how far away the person is.
What this means in practice
- •For robotics engineers: Detect if pedestrians are partially hidden in camera images without needing special labels or training data.
- •For autonomous vehicle developers: Improve pedestrian occlusion awareness in vehicle perception pipelines using existing monocular camera outputs without retraining occlusion models.
Authors
Chibueze Praise Onyenweaku, Toyibat Adele, Sikirulahi Opeyemi Abdulkareem, Obed Honour Eje
Abstract
Knowing that a pedestrian is occluded is useful to a robot even when the detector cannot recover the hidden pixels, but occlusion estimators are normally trained on occlusion annotations that deployment settings rarely have. We show that occlusion can be read off two distance estimates a camera-only pipeline already computes. Apparent height is largely unchanged by occlusion, because the detected box still spans the whole person; monocular metric depth is pulled toward the occluder, because the nearest surface inside the box is what a depth map reports. Only their ordering matters, so no focal length, stature, or depth scale is needed, provided the depth model's error is close to a pure gain, a condition that can be checked without labels. The score detects heavy occlusion on KITTI with AUC 0.909 (0.888 over all annotated pedestrians, including those the detector misses), requires no training and no occlusion labels, and is not a proxy for range: within fixed distance bands it holds AUC 0.894-0.925 while true distance alone sits near chance. On synthetic CARLA scenes with measured visibility it reaches 0.819, against 0.644 for distance alone. We arrived here after measuring three standard architectural responses to occlusion, depth fusion, occlusion-simulating augmentation, and attention, none of which improved detection outside its training distribution in the configurations we tested, and after finding that a detector trained on German roads loses half its accuracy on self-annotated street scenes in Lagos, Nigeria. Signals that need no local annotation are worth more in settings that benchmarks do not represent.