Convolutional Neural Network for License Plate Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Automatic Number Plate Recognition (ANPR) methods are error-prone due to the cumulative inaccuracies in character segmentation and recognition, leading to low accuracy in identifying license plates, especially with variations in size, position, and environmental conditions.
Innovation Solution
A method utilizing a convolutional neural network with separated symbol identification branches and a country identification branch, which analyzes the entire image without pre-segmentation, allowing for flexible recognition of license plates by identifying symbols and country-specific formats, improving precision and speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If character segmentation and independent recognition are used in the ANPR pipeline, then the recognition process can be structured and systematic, but the cumulative errors at each stage lead to low overall accuracy
Solution Approach 1:
The patent applies segmentation by dividing the license plate recognition task into multiple independent neural network branches, each responsible for identifying symbols at specific positions. This allows the system to process different regions of the license plate simultaneously through parallel computation, reducing cumulative errors while maintaining structured processing.
Solution Approach 2:
The patent merges multiple symbol identification branches into a single unified neural network architecture that processes the entire license plate image simultaneously. This integration eliminates the error propagation between separate processing stages while preserving the systematic approach to recognition.
2Productivity
If the entire image is analyzed without pre-segmentation using a convolutional neural network, then recognition speed and accuracy are improved, but the computational complexity increases
Solution Approach 1:
The patent segments the computational task into multiple specialized branches within the neural network, each handling specific symbol positions. This division allows parallel processing of different regions, improving computational efficiency and recognition speed while managing complexity through modular architecture.
Solution Approach 2:
The patent introduces a new dimensional organization to the computational process by arranging symbol identification branches in a structured spatial configuration that mirrors the license plate layout. This dimensional organization enables efficient parallel computation and reduces the overall computational burden.
3Measurement precision
If multiple separated symbol identification branches are used to identify symbols at respective positions, then recognition accuracy is enhanced, but the device structure becomes more complex
Solution Approach 1:
The patent creates a universal neural network architecture where multiple symbol identification branches share common computational resources and structural patterns. Each branch is designed to perform the same function (symbol identification) but for different positions, allowing the system to handle variable license plate formats through a single multi-functional network structure.
Solution Approach 2:
The patent manages structural complexity by dynamically adjusting parameters such as the number of branches, branch depths, and connection patterns based on the specific recognition task. This parameter optimization allows the network to achieve high accuracy while maintaining reasonable computational efficiency and structural simplicity.
4Adaptability or versatility
If country identification branch is separated from symbol identification branches, then country-specific license plate formats can be accurately recognized, but the overall system complexity increases
Solution Approach 1:
The patent segments the recognition system into functionally independent branches: symbol identification branches for character recognition and a separate country identification branch for determining the issuing country. This segmentation allows each branch to be optimized for its specific task while working together through the shared convolutional feature extraction layers, improving adaptability to different country formats.
Solution Approach 2:
The patent adds a new functional dimension to the recognition system by introducing the country identification branch that operates in parallel with symbol identification. This dimensional expansion enables the system to handle country-specific variations in license plate formats, symbols, and arrangements without compromising the core recognition functionality.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
The present invention provides an apparatus (100) for recognizing a license plate (105) of a vehicle (110), the apparatus (100) comprising an interface (140) for reading-in an image (135) of a surrounding of an optical sensor (130), the image (135) originating from the optical sensor (130) picturing at least said vehicle (110) having a detectable license plate (105). The apparatus (100) further comprises a unit (145) for analyzing the image (135) using a convolutional neural network (150), the convolutional neural network (150) having at least two separated symbol identification branches (220a, 220b), each of the separated symbol identification branches (220a, 200b) being configured for identifying one of several symbols (155) of the license plate (105) and/or the convolutional neural network (150) having a country identification branch (225) being separated from the symbol identification branches (220a, 220b), the country identification branch (225) being configured for identifying the country (160) having issued the license plate (105). Finally the apparatus (100) comprises a unit (165) for outputting (330) the at least two identified symbols (155) and/or the identified country (160) in order to recognize the license plate (105) of the vehicle (110).