automatic-number-plate-recognition（自动车牌识别）

TrafficEye License Plate Reader

Use this skill when the user wants to read a license plate from an image with the TrafficEye API.

What This Skill Does

1. Accepts a local image path.
2. Uploads the image to the TrafficEye recognition API.
3. Optionally sends a request form field if TRAFFICEYE_REQUEST_JSON is configured.
4. Parses the API response.
5. Picks the largest detected plate by polygon area.
6. Returns the full selected plate payload to the user, including text, type (country), dimension, scores, occlusion, unreadable, and position.

Expected Input

• A local image file path.
• If the user supplied an attachment instead of a path, first resolve it to a local file path and then run the helper.

Default Runtime Assumptions

• The API endpoint defaults to https://trafficeye.ai/recognition.
• The default request payload is {"tasks":["DETECTION","OCR"],"requestedDetectionTypes":["BOX","PLATE"]}.
• The default API-key transport matches the TrafficEye public API example: header mode with header name apikey.
• Auth and request fields remain configurable in case your deployment differs.

Environment Variables

• TRAFFICEYE_API_KEY: required unless passed explicitly to the helper.
• TRAFFICEYE_API_URL: optional, defaults to https://trafficeye.ai/recognition.
• TRAFFICEYE_API_KEY_MODE: one of header, bearer, form, query. Default: header.
• TRAFFICEYE_API_KEY_NAME: key name for header, form, or query mode. Default: apikey.
• TRAFFICEYE_FILE_FIELD: multipart field for the image. Default: file.
• TRAFFICEYE_REQUEST_FIELD: multipart field for the JSON request. Default: request.
• TRAFFICEYE_REQUEST_JSON: JSON string to include as the request field. By default this is {"tasks":["DETECTION","OCR"],"requestedDetectionTypes":["BOX","PLATE"]}.
• TRAFFICEYE_TIMEOUT_S: optional timeout in seconds. Default: 30.

How To Run

Setup your API key:

export TRAFFICEYE_API_KEY='YOUR_REAL_KEY'

Use the bundled helper:

python3 recognize_plate.py /absolute/path/to/image.jpg

For structured output:

python3 recognize_plate.py /absolute/path/to/image.jpg --format json

If the deployment expects Bearer auth:

TRAFFICEYE_API_KEY_MODE=bearer python3 recognize_plate.py /absolute/path/to/image.jpg

If the deployment needs an explicit request payload:

TRAFFICEYE_REQUEST_JSON='{"requestedDetectionTypes":["PLATE"]}' python3 recognize_plate.py /absolute/path/to/image.jpg --format json

Equivalent to the documented public API example:

curl -X POST \
  -H "Content-Type: multipart/form-data" \
  -H "apikey: YOUR_API_KEY_HERE" \
  -F "file=@image.jpg" \
  -F 'request={"tasks":["DETECTION","OCR"],"requestedDetectionTypes":["BOX","PLATE"]}' \
  https://trafficeye.ai/recognition

Agent Workflow

1. Verify that the image path exists.
2. Run python3 recognize_plate.py <image-path> --format json.
3. Present the full selected plate payload to the user, especially text, type, dimension, occlusion, unreadable, and position.
4. If the API returns no readable text, explain that the largest plate was found but OCR text was missing.
5. If authentication fails, ask the user which auth mode their deployment expects and retry with the matching environment variables.

Offline Validation

You can validate the selection logic without calling the API:

python3 recognize_plate.py --response-json-file examples/sample_response.json --format json

Notes

• The helper intentionally chooses the largest plate by geometric area, not by detection confidence.
• The response parser first checks combinations[].roadUsers[].plates[], then also supports roadUsers[].plates[], top-level plates[], and nested plate payloads discovered recursively.
• The default request and auth header mirror the public example at https://www.trafficeye.ai/api.
• The selected result now includes the original plate payload from the API so country/type and all scores are preserved.