License Plate Detection Using Temporal Consistency to Exclude Reflections
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing techniques face difficulties in accurately detecting license plates from images captured by digital cameras, particularly when images of license plates are reflected on glass, leading to erroneous detection.
Innovation Solution
An image processing method and apparatus that utilize machine learning to detect vehicle and license plate regions based on feature values like color and brightness, and employ a two-stage determination process to identify valid license plate regions by considering overlap and feature scores, thereby excluding erroneous detections.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple rectangles for license-plate-region candidates are detected from an input image, then the detection coverage is improved, but the number of false positives increases due to reflections on glass
Solution Approach 1:
The patent transitions from two-dimensional spatial detection in a single image to three-dimensional detection by incorporating temporal information across multiple continuous images. By tracking candidate regions through time and requiring consistency across frames, the system adds a temporal dimension that helps distinguish real license plates from transient reflections.
Solution Approach 2:
The system uses feedback mechanisms by continuously monitoring candidate regions across multiple frames. Detection results from previous frames inform the selection and validation of candidates in current frames, creating a feedback loop that refines detection accuracy and eliminates false positives through temporal consistency verification.
2Measurement precision
If character recognition and rectangle information are used to select license-plate candidates, then detection accuracy is improved, but processing time increases
Solution Approach 1:
The system performs preliminary detection of candidate regions in the first image before conducting detailed character recognition. By pre-identifying potential license plate regions and tracking them across frames, the system narrows down the search space early, reducing the overall processing time required for accurate detection.
Solution Approach 2:
The detection process is segmented into distinct stages: initial candidate region detection, temporal tracking across frames, character recognition, and final validation. This segmentation allows each stage to focus on specific tasks efficiently, reducing overall processing time while maintaining accuracy.
Data Source
AI summary
A method includes obtaining a plurality of images that are continuously captured, detecting a first region indicating a feature value corresponding to a license plate, in a region that indicates a feature value corresponding to a vehicle and that is included in a first image among the plurality of images, based on a feature value of each of the plurality of images, determining a second region in the plurality of images, the second region indicating a feature value corresponding to a license plate, at least a part of the second region overlapping the first region in an image different from the first image, and outputting the second region as a license plate region.