The sensor modality is in the claim, and here it is two of them. KAIST's grant US12651445B2, "System and method for perceiving 3-D environment using camera and radar" (issued June 9, 2026), is classified under G01S 13/867 — the CPC code for combining radar with data from another sensor — and G06V 20/58, recognition of objects relevant to vehicle operation. When a patent's classifications name both the radar-fusion code and the traffic-object code, you are looking at an automotive perception method, not a generic vision filing.
Why radar-and-camera rather than LiDAR? Cost and physics. Radar is cheap, works in rain and fog where cameras struggle, and measures velocity directly through Doppler shift — but it is coarse, with poor angular resolution. Cameras are rich in detail and color but estimate depth poorly and degrade in bad weather. Fusing the two aims to recover dense, weather-robust 3-D perception without the price of LiDAR. The claim is a bet that you can reach LiDAR-grade scene understanding with sensors a mass-market car can actually afford.
This is the kind of filing that makes a CPC class climb. The perception buckets — G06V 20/58, G01S 13/931, B60W 2420 — have been filling for years, but the composition is shifting from LiDAR-only methods toward fusion methods that mix cheaper sensors. A single university grant does not move a class on its own; it is a representative data point in a trend that the velocity counts confirm.
Note the assignee. This is the Korea Advanced Institute of Science and Technology — a university, not an automaker or a chip vendor. Academic assignees in the automotive perception classes are worth watching because they often license broadly, which means a method like this can end up in multiple companies' stacks rather than one. The IP-strategy read is different for a university grant than for a captive corporate one.
What the grant claims, precisely, is a method of perceiving a 3-D environment by combining camera and radar data — not radar, not cameras, not 3-D perception in the abstract. The scope is the fusion technique. That precision matters: prior art in radar-camera fusion is substantial, so the defensible novelty is in the specific way this method aligns and combines the two modalities, which the dependent claims pin down.
For an IP analyst, the takeaway is directional. The autonomy industry's quiet pivot from "LiDAR or nothing" toward affordable fusion stacks shows up first in the patent record, in classifications exactly like the ones on this grant. Watch G01S 13/867 — the radar-fusion code — as a leading indicator of where mass-market ADAS perception is heading.