Neural Network License Plate Recognition Without Character OCR
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automatic license plate recognition (ALPR) systems suffer from high error probabilities in recognizing license plate numbers, especially under adverse conditions, leading to the need for manual surveillance and delays in time-critical applications.
Innovation Solution
A method using an artificial neural network (NN) with one output node per unique license plate number, trained on a set of images to recognize license plates with low error rates, and extended by adding nodes for new plate numbers as needed, allowing for reliable recognition without separate character OCR and minimizing storage requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional ALPR schemes with separate character OCR are used, then the system can handle diverse license plate formats, but the error probability increases under adverse conditions
Solution Approach 1:
The patent merges the recognition of individual characters into a holistic recognition of the entire license plate number. Instead of processing each character separately through OCR, the system treats the complete license plate number as a single recognition unit, combining all character recognition steps into one unified process that reduces cumulative errors.
Solution Approach 2:
The system creates synthetic training images by copying and transforming existing license plate images. These synthetic copies are used to train the neural network, enabling it to learn from a expanded dataset without requiring additional real-world images, thus improving reliability under various adverse conditions.
2Reliability
If manual review by surveillance staff is implemented, then recognition reliability improves, but time consumption and costs increase
Solution Approach 1:
The system implements self-service by enabling the neural network to automatically recognize and verify license plate numbers without requiring manual intervention. The network independently processes images, identifies plate numbers, and handles recognition tasks, eliminating the need for surveillance staff to review each image manually.
Solution Approach 2:
The patent replaces the mechanical system of manual human review with an automated neural network system. The neural network performs the recognition function that previously required human operators, substituting automated computational processing for manual visual inspection, thereby reducing time loss while maintaining or improving reliability.
3Reliability
If more surveillance staff are employed, then recognition errors are reduced, but operational costs increase
Solution Approach 1:
The neural network system performs self-service by automatically detecting and correcting recognition errors without human intervention. The system independently processes images, validates recognition results, and handles edge cases, eliminating the need for additional surveillance staff to verify and correct errors.
Solution Approach 2:
The patent substitutes the mechanical system of human surveillance staff with an automated neural network. The network performs all recognition and verification functions that previously required human operators, replacing human labor with computational processing, thereby reducing the quantity of staff needed while maintaining high reliability.
4Reliability
If complete license plate numbers are recognized holistically, then error rates decrease, but the complexity of the recognition system increases
Solution Approach 1:
The neural network is designed with universality to handle multiple aspects of license plate recognition simultaneously. A single holistic recognition process manages the entire license plate number identification, replacing multiple separate character recognition systems, thereby reducing overall system complexity despite the advanced functionality.
Data Source
Figure 1
Figure 2~3
Figure 4~5
AI summary
A method (42) for automatic license plate recognition, ALPR, comprises recording (43) a first set (S1) of images (9i) each including one of N1 different license plate numbers (Lj); extracting (44) the license plate number (Lj) in each image (9i) of the first set (S1); generating (45) an artificial neural network (15) with one output node (Oj) for each of the N1 different license plate numbers (L1 - LN1); training (46) the artificial neural network (15) on the images (9i) and the extracted license plate numbers (Lj) of the first set (S1); recording (47) a sample image (9s); and feeding (48) the sample image (9s) into the artificial neural network (15) and recognising (49) the license plate number (Lj) of the sample image (9s) as the license plate number (Lj) of that output node (Oj) which outputs the highest value (27j). A system (1) carries out said method and a computing device (3) steps thereof.