Key Takeaways
- Why RFID beats computer vision on cost
- Checkout gate: full-bag scan in 0.5s
- Smart shelves and real-time inventory
01Why RFID beats computer vision on cost
Amazon Go's "Just Walk Out" relies on dozens of cameras plus edge GPUs, with a very high per-store cost — only viable for large stores. RFID costs a low per-tag price + one RFID gate + one edge middleware, totalling about 1/10 of computer vision. For 30-100 m² convenience stores, smart shelves and self-serve coffee, RFID is the only economically viable solution.
- Computer vision: very high per-store cost
- RFID: ~1/10 of computer vision cost
- Accuracy: RFID 99.9% vs CV 95-97%
- Best fit: RFID for 30-100 m² small stores
02Checkout gate: full-bag scan in 0.5s
RFID unmanned-retail checkout is dead simple: pick items, put them in a bag, walk to the exit gate. The reader identifies every EPC in the bag in 0.5s, auto-charges the bound account and opens the gate. No scanning, no queue, no clerk. An unmanned chain cut per-transaction time from 90s (traditional) to 8s, and night-time (22:00-6:00) sales rose from 5% to 22% — RFID makes 24/7 unmanned operation economically viable.
The engineering challenge of the checkout gate is miss-read and mis-read. Miss-read (customer took but didn't read) means direct loss; mis-read (read another customer's item) causes disputes. The solution is multi-antenna array + RSSI threshold filtering + secondary verification (face/QR scan at entry records item count, compared at exit). One vendor cut miss-read from 0.8% to 0.02% and mis-read to near zero with this combo.
03Smart shelves and real-time inventory
Another unmanned-retail pain is replenishment — unmanned, a stockout equals lost sales. RFID smart shelves (embedded antennas per layer) monitor item count in real time, auto-triggering replenishment work orders below a threshold and predicting replenishment timing based on historical sales velocity. An operator cut stockout hours from 8/day to 0.5/day, lifting per-shelf daily sales 22%. Smart shelves also detect "pick-and-return" behaviour to optimise display and pricing.
04Anti-theft and anomaly detection
Unmanned-retail anti-theft logic differs from traditional — no on-site clerk, so it must rely on tech. RFID provides three theft signals: 1) exit item count ≠ charged count; 2) shielding (foil wrapping blocks RFID) — the reader detects a sudden EPC disappearance without checkout; 3) abnormal dwell (customer stands long at a shelf while items decrease). One unmanned store combined RFID + cameras + behaviour analytics, cutting theft loss from 2.5% to 0.3%.
