LPR Preprocessing & Deskewing Pipeline

#Work

• Integrate LPR Preprocessing into InferServer ➕ 2026-07-07 🆔 7 ⛔ 5 Set up and migrate the license plate recognition (LPR) preprocessing logic (resizing, color space conversion, normalization) to run efficiently inside the InferServer environment.

  • Define the preprocessing pipeline configuration for InferServer.
  • Implement the preprocessing logic using InferServer's API.
  • Verify output consistency against the legacy preprocessing script.

• LPR Dataset Processing 🆔 we7ae1 ⛔ 5

  • Re-align dataset labels
  • train an OCR YOLO model

• Implement Edge Detection for Plate Deskewing 🆔 5 Investigate and implement traditional computer vision algorithms (e.g., Canny edge detection, Hough Transform, or Radon Transform) to detect plate angles and correct skew.

  • Create a prototype script using OpenCV for edge detection and affine transformations.
  • Test the algorithm on a sample batch of skewed/tilted plate images.
  • Document edge cases where classical methods fail (e.g., low contrast, dirty plates). Edge cases are for dark images and can be solved by adjusting the clip limit & tile grid size in CLAHE
  • Fix motion Blur and Apply sharpness to images.
  • Convert images to sharpen the images more.