UWB Plus Ceiling Cameras: Sub-30 cm Forklift Tracking, Fewer Tags
Tag the 25 forklifts, not the 20,000 pallets: how a UWB plus ceiling camera stack holds sub-30 cm accuracy with roughly half the anchors.
Hayat AminPresident of IP, Position Imaging
4 min read
Put one hardwired UWB tag on each forklift and let calibrated ceiling cameras carry the track between radio fixes. The tag gives identity and absolute position at 1 to 10 Hz, the cameras add 15 to 30 fps continuity, and a per-vehicle filter fuses both to hold sub-30 cm accuracy with roughly half the anchors of a radio-only design. Pallets need no tags at all: a pallet's location is the forklift's location at the moment of the drop.
Key takeaways
- A few dozen vehicle tags replace 20,000 pallet tags; a pallet's position equals the forklift's position at drop time.
- 802.15.4z UWB ranges to 10 to 30 cm with line of sight but degrades to meters behind steel racking; cameras fill the aisles.
- Fusion lets you place anchors at cross-aisles and docks only, cutting a 40 to 60 anchor radio-only plan to roughly 16 to 24.
- Sync cameras and anchors with PTP to sub-millisecond; at 3 m/s, 100 ms of clock error is your entire 30 cm accuracy budget.
- Ceiling cameras double as proximity-alert and dwell-time sensors, so the same hardware pays for two use cases.
- RF plus optical tracking fusion is patented ground; licensing granted IP beats 12 to 24 engineer-months of clean-room filter work.
Why tag the forklifts instead of the pallets?
A 200,000 square foot distribution center holds 15,000 to 30,000 pallet positions and runs 20 to 40 forklifts. Tag every pallet and you buy tens of thousands of UWB tags at 15 to 40 dollars each, then commit to a rolling battery-swap program measured in person-weeks per year. Tag only the forklifts and the hardware line item drops to a few dozen devices, each wired into vehicle power so batteries never enter the maintenance plan.
The insight that makes this work: pallets do not move themselves. If you know where every forklift is to 30 cm, and the WMS tells you which pallet each truck picked up, then put-away position equals forklift position at the moment of the drop. The vehicle becomes a roving measurement probe for the entire inventory.
- A few dozen vehicle tags replace 20,000 asset tags
- Vehicle power kills the battery replacement program
- Pallet location = forklift location at drop time
Tag the movers, infer everything else.
What do ceiling cameras fix that UWB cannot?
UWB under 802.15.4z uses roughly 500 MHz of bandwidth, which resolves the first-arriving signal path to 10 to 30 cm when the tag can see the anchor. Steel racking breaks that assumption. Behind a loaded rack the direct path is attenuated, the receiver locks onto a reflection, and a 20 cm fix quietly becomes a 1 to 3 m fix. A position solution also needs three to four anchors in view at once, which deep inside an aisle is exactly where you do not have them.
Ceiling cameras have the opposite failure mode. Mounted at 9 to 12 m and calibrated to the floor plane, they localize a forklift to well under 30 cm every frame, 15 to 30 times per second, with no tag in sight. What they cannot do is tell two identical reach trucks apart after a crossing or an occlusion.
Fused, each sensor covers the other's blind spot: the UWB fix binds a camera track to a specific vehicle ID and corrects drift; the camera carries that identity continuously between fixes. The radio names the track, cameras carry it.
How many anchors does fusion actually save?
Run the arithmetic on a 10-aisle facility with 100 m aisles. Radio-only, every fix needs line of sight to three or four anchors, and racked aisles are RF canyons, so you end up placing anchors every 25 to 35 m along each aisle. That is 40 to 60 anchors, each needing power, backhaul, and tight time sync, and anchor installation is usually the largest line in an RTLS quote.
A fused design places anchors only where geometry is good: cross-aisles, dock doors, staging areas. Inside the aisle, the camera carries the track and the filter re-anchors at every intersection. The same facility lands around 16 to 24 anchors plus roughly one camera per 15 by 25 m of floor.
The cameras are not a pure added cost. The same feeds drive pedestrian proximity alerts and dwell-time analytics, so one sensor serves two budgets.
Your exact counts depend on rack height and layout, so model propagation before you commit a bill of materials. Fewer anchors, and each camera earns twice.
What does the fusion pipeline look like in practice?
Four pieces, in order of how often they get botched:
- Time sync. Camera frames and UWB fixes must share a clock. A forklift at 3 m/s covers 30 cm in 100 ms, so 100 ms of timestamp error spends your entire accuracy budget. Use PTP and hold sync under a millisecond.
- Calibration. Survey each camera's extrinsics into the shared floor frame with ground targets, and re-verify after any mount gets bumped. A half-degree tilt error at a 10 m mounting height moves the ground-plane estimate by several centimeters.
- Estimation. Run one filter per vehicle, an EKF or factor graph with a constant-velocity-plus-yaw model. Camera detections update at 15 to 30 fps; UWB fixes arrive at 1 to 10 Hz, correct accumulated drift, and bind the track to a tag ID.
- Association. Gate camera detections against predicted positions. When two tracks merge under occlusion, the next UWB fix resolves which is which.
Budget under 100 ms end to end if the output feeds proximity alerts rather than dashboards. Sync the clocks first, everything else follows.
Should you build the fusion stack or license it?
The estimator above looks like a semester project and behaves like a multi-year one. Multipath rejection near metal racking, identity handoff across camera boundaries, calibration drift detection, and graceful degradation when a camera goes dark are each their own field-tuning campaign. Teams that build this clean-room typically spend 12 to 24 engineer-months before the system survives a real peak season, and they still face freedom-to-operate questions, because fusing RF ranging with optical tracking is well-covered patent ground.
That is the case for licensing. Position Imaging's portfolio spans real-time positioning, RF ranging, and computer vision tracking, includes granted US patents in this specific fusion space such as US 11,774,249 and US 12,079,006, and is cited by Apple and Bosch in their own filings. Licensing proven claims means your engineers spend their time on your product's workflow, not on re-deriving a tracking filter that already exists, and your counsel signs off before launch instead of after a demand letter.
Months to ship beats years to litigate.
Frequently asked questions
Can I track forklifts with no tags at all, cameras only?
You can hold position with cameras alone, but identity is fragile. Two identical trucks crossing paths or disappearing behind high stacking will swap IDs, and every swap corrupts your inventory record downstream. One hardwired UWB tag per vehicle is a small cost that makes identity errors self-correcting at the next radio fix.
What accuracy can UWB camera fusion realistically hold in a racked warehouse?
With sub-millisecond time sync and surveyed camera calibration, 15 to 30 cm is achievable across covered floor area. Accuracy degrades in coverage gaps where neither sensor sees the vehicle, so the design work is mostly about placing anchors and cameras so those gaps are short enough for the motion model to coast through.
How many UWB anchors do I need per square foot?
There is no flat density rule, because line of sight matters more than area. Radio-only designs in racked aisles need anchors every 25 to 35 m per aisle; a fused design needs them only at cross-aisles, docks, and staging, often cutting the count roughly in half. Model RF propagation against your actual rack layout before quoting hardware.
Do ceiling cameras create a driver privacy problem?
Less than you would expect. At 9 to 12 m mounting height with a top-down view, the system resolves vehicles, not faces. Most deployments run detection at the edge, keep only track coordinates with timestamps, and discard the raw frames, which also cuts storage and bandwidth. Confirm retention policy with your counsel and works council early.
How long does a retrofit take in a live facility?
Anchor and camera installation runs during off-shifts, so operations do not stop. Plan a single pilot aisle first to validate calibration and sync in your building, then scale. The long poles are usually cabling and ceiling access lifts, not the software, so get your electrical contractor into the schedule before the hardware arrives.
Send us your floor plan and accuracy target, and we will map your forklift tracking design to the specific granted patents it touches.
Tell us the product. We map the exact scope, what a license covers, and how fast you can ship, all in a 20-minute call.
Book a 20-minute call