License Plate Detection Recognition Edge Device Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing License Plate Detection and Recognition (LPDR) systems face accuracy issues in complex situations such as shadows, noise, and poor visibility, and require powerful hardware for real-time processing, making them unsuitable for edge devices without GPUs.
Innovation Solution
A LPDR system using a combination of deep learning neural networks for license plate detection and optical character recognition, allowing separate processing of detection and recognition tasks, and capable of handling partially visible, stacked, and low-contrast license plates, optimized for performance on less powerful hardware.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If computer vision-based techniques are used for license plate detection and recognition, then the system can operate in simple situations with clear visibility, but the accuracy is compromised in complex situations with shadows, noise, and dust
Solution Approach 1:
The patent transforms the license plate image from its original orientation to a horizontally aligned state by detecting the plate's angle and applying geometric transformation. This parameter change (orientation alignment) enables the recognition system to accurately process plates regardless of their initial angular position, thereby improving measurement precision in complex situations while maintaining reliability in simple cases
2Measurement precision
If modern real-time license recognition models are deployed, then recognition accuracy is improved, but computational cost increases requiring powerful hardware such as GPU
Solution Approach 1:
The patent divides the license plate recognition task into two independent modules: a detection module that locates the license plate region and determines its orientation, and a recognition module that performs character recognition on the aligned plate image. This segmentation allows each module to be optimized separately, reducing overall computational cost while maintaining high accuracy, making the system suitable for edge devices with limited resources
3Use of energy by stationary object
If real-time license recognition is performed on edge devices without GPU, then power consumption is reduced, but the system lacks the computational capacity for modern recognition models
Solution Approach 1:
The patent extracts and removes the computationally intensive orientation detection and alignment operations from the recognition pipeline, performing them as preliminary preprocessing steps. By separating these geometric transformations from the core neural network recognition process, the system reduces the computational burden on edge devices while maintaining recognition accuracy, enabling deployment on devices with limited processor capacity
Data Source
AI summary
A license plate detection and recognition system receives training data comprising images of license plates. The system prepares ground truth data from the training data based predefined parameters. The system trains a first machine learning algorithm based on the ground truth data to generate a license plate detection model. The license plate detection model is configured to detect one or more regions in the images. The one or more regions contains a candidate for a license plate. The LPDR system generates a bounding box for each region. The LPDR system trains a second machine learning algorithm based on the ground truth data and the license plate detection model to generate a license plate recognition model. The license plate recognition model generates a sequence of alphanumeric characters with a level of recognition confidence for the sequence.