Researchers led by Soham Pahari introduced CoCam4D, a camera‑only collaborative perception system for autonomous driving. It models geometric uncertainty with 3D Gaussian scene primitives and shares them via C‑V2X using a 35‑byte format called Dynamic Object Primitives. Tests on public datasets show an 11.48% boost over the best recent vision‑only method on OPV2V+ and a 10.62% gain on DAIR‑V2X‑C. The approach works without LiDAR, relying only on shared camera data.
Why it matters
It lets fleets of camera‑equipped cars improve perception in blind spots without costly LiDAR hardware.
