Tencent Cloud License Plate OCR

The Problem

Vehicle images often contain parked cars, background clutter, glare, and multiple plates. Manual entry is slow and error-prone, while generic OCR may not return plate color, category, or bounding boxes in a usable form. @tencent-adm/tencentcloud-ocr-licenseplate targets mainland China vehicle plates: extract plates from one image and return the plate number, color, category, and pixel coordinates.

How It Works

The skill wraps Tencent Cloud OCR's LicensePlateOCR API. Provide either ImageBase64 or ImageUrl; if both are present, ImageUrl takes priority. The service downloads and decodes the image, detects plates, and returns structured plate results:

• Number: plate number, e.g. 京AF0236
• Color: white, black, blue, green, yellow, yellow-green, temporary, sprayed, or other
• LicensePlateCategory: physical or non-physical plate
• Rect: pixel box with X, Y, Width, Height
• LicensePlateInfos: full list when multiple plates exist
• PlateCount: detected plate count

For engineering workflows, the result can feed registration, toll or parking reconciliation, fleet tracking, or a human review UI. When multiple plates are present, prefer LicensePlateInfos over single top-level fields.

Boundaries and Notes

• Supports PNG/JPG/JPEG; GIF is not supported
• ImageBase64 payload should be under 10M, and ImageUrl downloads should complete within 3 seconds
• Default request rate limit is 10 requests per second
• Common errors include download failure, image decode failure, OCR failure, service not enabled, and exhausted resource packages
• Best suited for mainland China plates; overseas plates, handwritten plates, or heavily occluded plates may need human review