LPR Data Classes

Overview

License plate data in degirum-vehicle is represented by two related classes:

• PlateOCRResult - Contains OCR data (plate number, confidence)

• LPRResult - Includes all PlateOCRResult properties plus detection metadata (bounding box, detection score)

PlateOCRResult

Contains license plate OCR data without detection-specific information. Used primarily for tracking workflows.

Properties

Property	Type	Description
plate_number	str or None	Recognized license plate text
ocr_score	float	OCR confidence score 0.0-1.0

When It's Used

find_plates_in_file() and find_plates_in_clip() return Dict[int, PlateOCRResult] - a mapping of track IDs to plate data. Each PlateOCRResult object contains:

• Aggregated plate text - Fused across frames using Bayesian text aggregation

• Confidence score - Final OCR confidence after multi-frame fusion

• This accumulated data across frames provides robust plate recognition results

Usage Example

# Analyze video to extract license plates
plates = tracker.find_plates_in_file("parking_lot.mp4")

# plates is Dict[int, PlateOCRResult] - maps track_id to plate data
for track_id, plate in plates.items():
    print(f"Track {track_id}:")
    print(f"  Plate: {plate.plate_number}")
    print(f"  OCR Score: {plate.ocr_score:.3f}")

LPRResult

Subclasses PlateOCRResult to add detection-specific metadata. Used for real-time recognition results.

Properties

Inherits all PlateOCRResult properties (plate_number, ocr_score) plus:

Property	Type	Description
bbox	list	Bounding box [x1, y1, x2, y2]
detection_score	float or None	Plate detection confidence 0.0-1.0

Important notes:

• When using LicensePlateTracker, track_id is in InferenceResults.results, not as a property. Access via result.results[i].get("track_id")

When It's Used

• Contained in InferenceResults.license_plates from predict() and predict_batch()

Usage Examples

Access detection properties:

result = recognizer.predict("car.jpg")

for plate in result.license_plates:
    if plate.plate_number:
        print(f"Plate: {plate.plate_number} (OCR: {plate.ocr_score:.2f})")
    else:
        print(f"Unreadable plate (detection: {plate.detection_score:.2f})")
    
    # Access bounding box
    x1, y1, x2, y2 = plate.bbox
    print(f"Plate location: ({x1}, {y1}) to ({x2}, {y2})")

LicensePlateTracker - Access tracking data:

for result in tracker.predict_batch(video_source):
    # result.license_plates and result.results correspond at the same index
    for i in range(len(result.license_plates)):
        track_id = result.results[i].get("track_id")
        plate = result.license_plates[i]
        print(f"Track {track_id}: {plate.plate_number} (OCR: {plate.ocr_score:.2f})")

String representation:

for result in lpr.predict_batch(["car1.jpg", "car2.jpg"]):
    for plate in result.license_plates:
        # Pretty print using __str__
        print(plate)

Output:

Plate Number    : ABC1234
OCR Score       : 0.987
Detection Score : 0.952
Bounding box    : [123, 456, 789, 234]

Return Types by Method

LicensePlateRecognizer

Method	Return Type
predict()	InferenceResults (.license_plates property yields LPRResult objects)
predict_batch()	Iterator of InferenceResults

LicensePlateTracker

Method	Return Type
predict_batch()	Iterator of InferenceResults (.license_plates property yields LPRResult objects)
find_plates_in_file()	Dict[int, PlateOCRResult]
find_plates_in_clip()	Dict[int, PlateOCRResult]