Automated Image Rejection for License Plate Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current License Plate Recognition (LPR) systems are inefficient in accurately identifying the absence of a license plate in images, leading to unnecessary human reviews and associated costs, due to high false alarm rates and rejection of valid images, which results in lost revenue and increased expenses.
Innovation Solution
A method and system that utilize a no-plate classifier trained on local and global image features to determine the likelihood of a license plate being absent, obstructed, or unreadable, with a confidence value compared against a threshold to automate the image rejection process, reducing human review requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If LPR systems use highly selective rejection criteria to ensure high OCR accuracy (99% or better), then the accuracy of license plate recognition is improved, but the number of false positives increases and more images require human review
Solution Approach 1:
The system segments the license plate detection process into multiple independent stages: initial detection, candidate generation, feature extraction, confidence scoring, and final verification. Each stage applies specific filtering criteria, allowing the system to maintain high accuracy while reducing false positives by progressively eliminating uncertain candidates before human review is needed
Solution Approach 2:
The system dynamically adjusts detection parameters such as confidence thresholds, ROI size, and character segmentation criteria based on image quality metrics. By changing these parameters adaptively, the system optimizes the balance between catching all valid plates and avoiding false positives, reducing the burden on human reviewers
2Reliability
If LPR systems generate many candidate ROI's to ensure license plate detection, then the reliability of detecting actual license plates is improved, but the processing time and computational resources increase
Solution Approach 1:
The system performs preliminary filtering operations before full OCR processing, including initial plate localization, aspect ratio validation, and basic character count verification. This preliminary action eliminates obviously invalid candidates early in the pipeline, allowing the system to maintain high detection reliability while significantly reducing the number of images that require full processing time
Solution Approach 2:
The system applies partial processing to candidate images based on their confidence scores. High-confidence candidates receive full OCR processing, while low-confidence candidates receive only basic validation or are flagged for human review. This partial action approach maintains detection reliability for clear cases while reducing overall processing time by avoiding exhaustive analysis of uncertain cases
3Measurement precision
If LPR systems forward images with low confidence conclusions to human review, then the accuracy of toll collection is improved, but the cost and time for human review increases
Solution Approach 1:
The system implements feedback loops where human review outcomes are fed back into the training dataset. This allows the system to learn from actual human decisions and continuously improve its confidence scoring. As a result, the system progressively reduces the number of images requiring human review while maintaining or improving accuracy, thereby reducing the energy cost of human intervention
Solution Approach 2:
The system replaces manual human review with an automated confidence scoring mechanism that uses machine learning models to predict the likelihood of correct recognition. This mechanical substitution handles the majority of clear cases automatically, reserving human review only for ambiguous cases where the automated system's confidence falls below a threshold, thereby significantly reducing human review costs while maintaining accuracy
Data Source
AI summary
Systems and methods for automating an image rejection process. Features including texture, spatial structure, and image quality characteristics can be extracted from one or more images to train a classifier. Features can be calculated with respect to a test image for submission of the features to the classifier, given an operating point corresponding to a desired false positive rate. One or more inputs can be generated from the classifier as a confidence value corresponding to a likelihood of, for example: a license plate being absent in the image, the license plate being unreadable, or the license plate being obstructed. The confidence value can be compared against a threshold to determine if the image(s) should be removed from a human review pipeline, thereby reducing images requiring human review.