License Plate Recognition via Multi-Stage CNN and Tensor Decomposition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Convolutional neural networks (CNNs) are inefficient for mobile devices due to high memory and computational complexity, making them unsuitable for accurate license plate recognition without significant size reduction.
Innovation Solution
A multi-stage license plate recognition method using a vehicle searching CNN, license plate searching CNN, tensor neural network for anomaly detection, and separate CNNs for area number, letter, and number recognition, with space transformation networks and tensor decomposition to optimize network size and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a convolutional neural network with larger size of adjustable parameters is used to improve image recognition accuracy, then classification accuracy is improved, but memory complexity and computational complexity increase, making it unsuitable for mobile devices
Solution Approach 1:
The patent segments the license plate recognition task into multiple independent stages: vehicle detection, license plate localization, character segmentation, and character recognition. Each stage uses a dedicated neural network model with optimized parameters. This segmentation allows each sub-network to be smaller and more efficient while collectively achieving high accuracy, resolving the contradiction between network size and classification accuracy.
2Productivity
If distributed parallel computing acceleration methods are used to improve processing speed, then computational efficiency is improved, but device complexity increases and becomes impractical for mobile devices
Solution Approach 1:
The patent employs dynamic optimization techniques where neural network parameters and processing configurations are adaptively adjusted based on input characteristics and device capabilities. The system dynamically selects appropriate processing paths and optimizes computational resources in real-time, achieving high processing speed without requiring complex fixed parallel computing architectures.
Data Source
AI summary
The present disclosure provides a license plate recognition method, a device thereof, and a user equipment. In the method, a vehicle image is acquired and input to a vehicle searching convolutional neural network (CNN) to search whether there is a vehicle. Then, a license plate image is acquired and input to a license plate searching CNN to search whether there is a license plate. And the license plate image is input to a tensor neural network to detect abnormal or not. Then, a recognition image is acquired and input to judge an area number, letters, and numbers. And an area number image, a letter image, and a number image are output and input to their corresponding recognition CNNs respectively to recognize an area number, letters, and numbers and output respectively. Ultimately, a license plate number recognition result in a form of “area number letter number” are output.