Edge License Plate Recognition Using CIoU Neural Network
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing license plate recognition systems require high network traffic and expensive hardware, such as cameras with higher shutter speeds, due to insufficient processing capabilities on-site and the need for image transmission to servers, especially when dealing with blurry images from moving vehicles.
Innovation Solution
A method and device that perform real-time license plate number recognition using a neural network with a loss function comprising classification, confidence, and Complete Intersection over Union (CIoU) loss, running on low-resource IoT devices with on-board computations, eliminating the need for video frame transmission and high-speed cameras.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If photos are transmitted to a backend server for license plate recognition, then recognition can be performed, but network traffic occupation increases significantly
Solution Approach 1:
The patent extracts the license plate recognition functionality from the backend server and implements it locally on edge devices (set-top boxes, surveillance cameras, mobile devices). This extraction eliminates the need to transmit photos to servers, thereby reducing network traffic occupation while maintaining recognition accuracy through local processing
Solution Approach 2:
The patent introduces an intermediary layer - the edge device with embedded neural network - that processes images locally before any potential server communication. This intermediary handles the recognition task independently, serving as a mediator that prevents direct photo transmission to servers and reduces network bandwidth consumption
2Measurement precision
If better cameras with higher shutter speed are installed to capture clear images of moving vehicles, then image quality improves, but hardware deployment cost increases
Solution Approach 1:
The patent changes the parameter of image processing by introducing advanced neural network algorithms that can recognize license plates in blurry images. Instead of improving camera hardware parameters (shutter speed), the solution modifies the software processing parameters to extract meaningful information from lower-quality images, thereby avoiding expensive hardware upgrades
Solution Approach 2:
The patent replaces the mechanical/optical solution (higher shutter speed cameras) with an computational/intelligence-based solution (neural network processing). The neural network compensates for motion blur and image quality issues through algorithmic enhancement, substituting hardware improvements with software intelligence
3Loss of energy
If on-board computations are performed on low-resource IoT devices, then network bandwidth and hardware costs are reduced, but processing capability constraints increase
Solution Approach 1:
The patent segments the license plate recognition task into distinct processing stages (preprocessing, feature extraction, recognition, post-processing) that can be distributed across different components of the edge device. This segmentation allows efficient utilization of limited computational resources by allocating specific tasks to appropriate hardware units within the constrained device
Solution Approach 2:
The patent uses optimized copies of neural network models that are specifically designed for deployment on resource-constrained devices. These copied models are pruned, quantized, or otherwise adapted to run efficiently on low-power hardware while maintaining acceptable recognition accuracy
Data Source
AI summary
The present disclosure proposes a method for recognizing a license plate number on an image, comprising the following steps: detecting a license plate on a vehicle in an image; recognizing characters on the license plate and coordinates of the characters by a neural network, wherein the loss function of the neural network consists of classification loss, confidence loss and Complete Intersection over Union (CIOU) loss; and organizing the recognized characters based on the coordinates to form the recognized license plate number. By this method, when recognition of a license plate number on an image is performed, the network bandwidth and the hardware cost may be greatly reduced.