Edge ANPR & License Plate Recognition

Read license plates with the gate cameras you already have, and lift the barrier for whitelisted vehicles.

Read license plates from your existing gate cameras with one on-site recognition host, and open the barrier automatically for whitelisted vehicles. Every vehicle leaves a record with its plate number and a snapshot; plates and images stay on site.

Seeed LabIntermediate45minVision & Video AI

Plate preview on a reComputer J30 Series: blue plate 皖A62599 read at 0.98 from an oblique front view, with the last five reads on the right (CCPD test image, MIT licence)

reComputer J30 Series: green new-energy plate 皖AF10528 read at 0.87 from the front (CCPD test image, MIT licence)

reComputer RK3588 Series: blue plate 皖ALY625 read at 0.90 from the rear (CCPD test image, MIT licence)

Reusable Site Templates

Gates where vehicles pass through a fixed lane and the plate is legible in the picture.

Residential and campus gates

Import residents' plates into the whitelist and the barrier lifts on arrival; visitors still go through the guard booth.

Staff car parks

Import staff plates with an expiry date; access ends automatically when it expires.

Entry and exit records only

Without a barrier, record each vehicle's plate, time and snapshot for your parking system to reconcile.

Supports mainland China blue plates and green new-energy plates; two-row, special-purpose and overseas plates are not supported.

How well it works

What you see on the recognition host's browser preview page once it is running.

Live plate preview

  • The plate is boxed and labelled with its number and confidence
  • The last five reads are listed on the right

Blue and green new-energy plates

  • Reads 7-character blue and 8-character green plates
  • Two-row, special-purpose and overseas plates are not supported

One record per vehicle

  • Plate number, confidence and a snapshot link in one message
  • Snapshots stay on the recognition host for 7 days by default
Measured results

Input is a 1080p RTSP stream looping license-plate stills from the public CCPD dataset.

MetricTypicalDeviceDeployment Note
1080p processing speed30 fps (single-frame p50 5.6 ms)reComputer J30 Series (J3011)Default recommendation, TensorRT FP16 acceleration
1080p processing speed29–30 fps (single-frame p50 23.1 ms)reComputer RK3576 SeriesCost-effective choice, onboard NPU INT8 inference
Plate verification (8 stills)5 correct, 3 withheld, 0 wrong platesreComputer RK3588 SeriesRockchip flagship platform with multi-model headroom

Before going live, check the reads against real plates using daytime and night footage from your own gate. For per-device latency and benchmark matrices, see Engineering Wiki · Edge ANPR Solution.

How to deploy

Check the setup and its product list, then follow the deployment guide.

Deployment Options

IP Camera + Recognition Host

Keeps your existing gate cameras. Choose a reComputer J30 / J40 (Jetson Orin Nano / Orin NX, JetPack 6.2), RK3588 or RK3576 recognition host; add a reComputer R1100 Series gate controller to open the barrier automatically.

4 steps · ~45min

Recognition Host
recomputer_j30
recomputer_j40
recomputer_rk3588
recomputer_rk3576
IP Camera
IP Camera
Gate Controller (optional)Optional
recomputer_r11

Integration Interfaces

mqtt

One record per vehicle — plate number, plate type, confidence, position in the image and a snapshot link. Published to the MQTT broker you set in Step 1 (usually the reComputer R1124-10).

Port: 1883 · Topic: <site>/parking/<camera>/plate
{"schema":"parking.plate/1","event_id":"...","site_id":"demo-lot","camera_id":"gate-a","ts":1790000000789,"seq":2211,"replayed":false,"plate":"粤B12345","plate_normalized":"粤B12345","region":"cn","plate_type":"cn_blue","confidence":0.93,"votes":5,"frames":7,"bbox":[0.41,0.62,0.55,0.67],"track_id":77,"snapshot":{"url":"http://camera.local:8080/snapshots/2211.jpg","sha256":"...","width":1920,"height":1080}}
mqtt

Gate records from the R1124-10 — action open_pulse (barrier opened) or suppressed (same plate again within the cooldown), with the matching plate record and whitelist label.

Port: 1883 · Topic: <site>/parking/<camera>/gate
{"schema":"parking.gate/1","event_id":"...","site_id":"demo-lot","camera_id":"gate-a-barrier","ts":1790000000912,"seq":18,"replayed":false,"action":"open_pulse","trigger":"whitelist","plate_event_id":"...","plate_normalized":"粤B12345","whitelist_label":"员工-张","pulse_ms":500,"latency_ms":123}
mqtt

Status of the recognition service, published periodically; the last status is kept on the broker if the host goes offline.

Port: 1883 · Topic: <root>/status
http

Snapshot linked from each plate record. Stored on the recognition host for 7 days by default; the image is never embedded in the record.

/snapshots/{seq}.jpg · Port: 8080 · Method: GET
http

Whitelist on the R1124-10 — GET returns the current whitelist CSV, PUT replaces it (a 400 response names the failing line). POST /gate/test fires one test opening.

/whitelist · Port: 8081 · Method: GET
http

Status check of the recognition service, including recognition speed.

/healthz · Port: 8099 · Method: GET

Usage Requirements

rtsp

A fixed view of the lane at 3–8 m where plates stay legible, delivered as an RTSP stream from your own IP camera (1080p or better).

Contact Us
We Are Glad to Be Your Hardware Partner !
Next