CN-121095929-B - Identification method for curb plate of curb locomotive

CN121095929BCN 121095929 BCN121095929 BCN 121095929BCN-121095929-B

Abstract

The invention discloses a recognition method of a curb plate, which comprises the steps of calculating a correlation coefficient between an environment variable and recognition performance by adopting a trend model according to an extracted feature vector to obtain a performance attenuation trend index, carrying out sensitivity calibration operation on a sensor array if the performance attenuation trend index exceeds a preset threshold value, obtaining an adjusted sensor response curve as a dynamic optimization intermediate result, determining an updating range of an algorithm threshold value from the adjusted sensor response curve, fusing environment variable data by an iterative optimization method to obtain an optimized threshold value set to adapt to the current dynamic environment, and updating intelligent system configuration by adopting the determined threshold value fine adjustment parameter to obtain an enhanced recognition module for processing an image data stream of a curb plate in a complex scene.

Inventors

CHEN XIAOMIAN
GAN YANG
LI BO
WANG LEI
TAO YU
LUO JIN

Assignees

四川易智停科技有限公司

Dates

Publication Date: 20260508
Application Date: 20250821

Claims (9)

1. A recognition method of a curb plate is characterized by comprising the following steps: S101, extracting a feature vector from environment variable data as an input basis of a trend model by collecting real-time environment variable data in a recognition process of a curb plate; S102, calculating a correlation coefficient between an environment variable and identification performance by adopting a trend model according to the extracted feature vector to obtain a performance attenuation trend index, wherein the performance attenuation trend index is used as a reference basis for subsequent parameter adjustment; S103, if the performance attenuation trend index exceeds a preset threshold, performing sensitivity calibration operation on the sensor array, and acquiring an adjusted sensor response curve as an intermediate result of dynamic optimization; s104, determining an updating range of an algorithm threshold from the adjusted sensor response curve, and fusing environment variable data by an iterative optimization method to obtain an optimized threshold set to adapt to the current dynamic environment, wherein the method comprises the following steps: acquiring original response data from a sensor, denoising and normalizing the data by adopting a signal processing method to obtain an adjusted response curve; Extracting key feature points through curve analysis according to the adjusted response curve, and determining an initial threshold range; Acquiring environment variable data, and if the environment variable data is not matched with a preset range, fusing environment variables through a weighted average method to obtain comprehensive environment parameters; According to the comprehensive environment parameters and the initial threshold range, performing iterative optimization by adopting a gradient descent algorithm to obtain an optimized threshold set; aiming at the optimized threshold value set, if the dynamic environment changes, the threshold value set is adjusted by monitoring and updating environment parameters in real time; judging whether the threshold value is suitable for the current dynamic environment or not by comparing the adjusted threshold value set with sensor response data to obtain a final threshold value set; Generating control parameters adapting to the dynamic environment according to the final threshold value set, and outputting the control parameters to a system execution module; s105, after the optimized threshold value set is obtained, judging the matching degree of the threshold value set and the identification performance, if the matching degree is lower than the standard value, re-analyzing trend model output through a feedback loop mechanism, and determining further threshold fine tuning parameters; S106, updating intelligent system configuration by adopting the determined threshold fine tuning parameters to obtain an enhanced identification module for processing the image data stream of the license plate of the curb machine in a complex scene; s107, acquiring the processed image data stream from the enhanced recognition module, judging the applicability of the processed image data stream in the intelligent parking scene, and outputting a final license plate recognition result to support traffic management application if the applicability meets the requirement.
2. The method for recognizing a guidepost according to claim 1, wherein in S101, extracting feature vectors from the environmental variable data as input basis of a trend model by collecting real-time environmental variable data in a guidepost recognition process comprises: acquiring environment variable data including illumination intensity and vehicle speed in real time through a sensor to obtain an original data set; denoising and normalizing the original data set by adopting a preprocessing method to obtain a normalized data set; Extracting feature vectors of illumination intensity and vehicle speed from the standardized data set to obtain a feature vector set; Judging the feature vector set, and if the illumination intensity in the feature vector set is lower than a preset threshold value, adjusting the image brightness by using an image enhancement algorithm to obtain an enhanced feature vector set; classifying the enhanced feature vector set through a support vector machine algorithm, and judging the input category of a trend model of license plate recognition; Adjusting trend model parameters according to the classification result to obtain an optimized trend model; And predicting the environmental variable data acquired in real time by adopting an optimized trend model to obtain a trend result of license plate recognition.
3. The method for recognizing a curb plate according to claim 2, wherein S102, calculating a correlation coefficient between the environment variable and the recognition performance by using a trend model according to the extracted feature vector to obtain a performance attenuation trend index comprises: normalizing the feature vector by adopting a preprocessing method from the acquired feature vector and the environment variable to obtain a standardized feature vector; Calculating the association coefficient between the standardized feature vector and the environment variable by adopting a Pearson correlation coefficient method to obtain the association strength of the environment variable and the identification performance; Judging the obtained association strength, and if the association strength exceeds a preset threshold, adopting a linear regression model to fit the relationship between the environment variable and the recognition performance to obtain trend model parameters; calculating performance attenuation trend indexes according to the trend model parameters, and determining performance change trend; Extracting key points from the performance change trend, and optimizing a parameter adjustment strategy by adopting a gradient descent method to obtain an adjusted parameter set; updating the configuration of the recognition model according to the adjusted parameter set to obtain the optimized recognition performance; And comparing the optimized identification performance with the original performance, and evaluating the parameter adjustment effect to obtain the performance improvement index.
4. The method for identifying a curb plate according to claim 3, wherein in S103, if the performance attenuation trend index exceeds a preset threshold, performing a sensitivity calibration operation on the sensor array, and obtaining an adjusted sensor response curve as an intermediate result of dynamic optimization, comprises: If the performance attenuation trend index exceeds a preset threshold, acquiring real-time signal data from the sensor array through a data acquisition module to obtain an original response data set; Filtering and denoising the original response data set through a signal processing technology to obtain a smooth response data set; Carrying out feature extraction on the smooth response data set by adopting a support vector machine algorithm, and determining key feature parameters of performance attenuation; If the deviation between the key characteristic parameter and the preset threshold exceeds a specified range, triggering a sensitivity calibration module to acquire a calibration parameter set; adjusting the sensitivity setting of the sensor array through the calibration parameter set to obtain an adjusted response curve; Performing fitting analysis on the adjusted response curve by adopting a linear regression algorithm to obtain a dynamic optimized intermediate result; and continuously tracking the dynamically optimized intermediate result through a real-time monitoring module, and judging whether the performance attenuation trend is recovered to be within a preset threshold value.
5. The method for identifying a curb plate according to claim 1, wherein in S105, after the optimized threshold set is obtained, the matching degree between the curb plate and the identification performance is determined, if the matching degree is lower than a standard value, the trend model output is re-analyzed through a feedback loop mechanism, and further threshold fine tuning parameters are determined, including: after the threshold value set is obtained, calculating the matching degree between the threshold value set and the identification performance through a support vector machine algorithm to obtain a matching degree value; If the matching degree value is lower than a preset threshold value, extracting trend data output by the model through a feedback loop mechanism to obtain a trend analysis result; Optimizing a threshold set by adopting a gradient descent algorithm according to the trend analysis result, and determining a fine tuning parameter set; adjusting the threshold value set through the fine adjustment parameter set to generate an updated threshold value set; Re-calculating the matching degree of the updated threshold value set and the identification performance to obtain a new matching degree value; If the new matching degree value is still lower than the preset threshold value, repeating the feedback loop and the parameter adjustment steps to obtain a final optimized threshold value set; And performing performance evaluation through the final optimized threshold set, and determining the stability of the system identification performance.
6. The method for recognizing a curb plate according to claim 5, wherein in S106, the intelligent system configuration is updated by using the determined threshold fine tuning parameters to obtain an enhanced recognition module for processing an image data stream of the curb plate in a complex scene, comprising: determining a parameter adjustment range through a preset threshold value, acquiring a curb plate image data stream from a complex scene, and generating an initial image set; If the initial image set contains noise, processing the initial image set by adopting a Gaussian filter algorithm to obtain a denoising image set; extracting features of the road-dental locomotive license plate by adopting a convolutional neural network according to the denoising image set to generate a feature image set; If the resolution ratio of the feature image set is lower than a preset threshold value, enhancing the feature image set through a super-resolution algorithm to obtain an enhanced image set; analyzing license plate characters by adopting an optical character recognition algorithm according to the enhanced image set to generate a character data stream; the character data stream is matched with a preset license plate template, and the matching consistency is judged to obtain a license plate recognition result; And updating the intelligent system configuration according to the license plate recognition result to generate an enhanced recognition module.
7. The method for recognizing a curb plate according to claim 1, wherein S107, obtaining the processed image data stream from the enhanced recognition module, judging the applicability of the processed image data stream in a smart parking scene, and if the applicability meets the requirement, outputting a final license plate recognition result to support traffic management application, comprises: acquiring an image data stream from an enhancement recognition module, denoising and enhancing the image by adopting a preprocessing algorithm to obtain first image data; according to the first image data, extracting license plate region features by adopting a convolutional neural network algorithm to obtain license plate positioning data; aiming at license plate positioning data, if the feature definition exceeds a preset threshold, analyzing license plate characters by adopting an optical character recognition algorithm to obtain license plate text data; judging whether the license plate text data meets the license plate format requirement of the intelligent parking scene according to the license plate text data to obtain a format verification result; if the format verification result meets the requirements, matching license plate text data with a preset parking lot vehicle database to obtain vehicle identity information; Generating a license plate recognition result according to the vehicle identity information and outputting the license plate recognition result to a traffic management application interface to obtain final recognition output; And updating the parking lot management database through final identification output to complete data synchronization of traffic management application.
8. The method for identifying a curb plate according to claim 7, further comprising a fusion identification process of the high dynamic range image stream, wherein the method specifically comprises the following steps: Acquiring multi-source sensor data, preprocessing and denoising to generate a first data stream; extracting light characteristics through real-time light change rate analysis to obtain light adjustment parameters; Fusing the first data stream based on the light ray adjustment parameters to generate a high dynamic range image stream; If the definition and data integrity of the image stream meet the preset threshold, extracting license plate features through a convolutional neural network, and determining a license plate character sequence by combining an optical character recognition algorithm; and comparing the character sequence with a preset database, and outputting a final license plate recognition result.
9. The method for identifying a curb plate according to claim 8, further comprising an optimized identification process of a high dynamic range image stream, specifically comprising: Acquiring original image data through a multi-source sensor, and generating a high dynamic range image stream by adopting a data fusion technology; if the definition of the image stream is lower than a preset threshold value, optimizing the contrast and the brightness through an image enhancement algorithm to obtain an optimized image stream; Analyzing and optimizing the light change rate of the image stream, adjusting, and extracting license plate region features through a convolutional neural network if the adjusted data integrity meets a threshold value; analyzing the character sequence by adopting an optical character recognition algorithm, comparing a preset license plate format database to verify the legality of the character sequence, and outputting a license plate recognition result.

Description

Identification method for curb plate of curb locomotive Technical Field The invention belongs to the technical field of image recognition, and particularly relates to a recognition method for a curb plate. Background In an intelligent parking scene, the curb plate recognition system needs to deal with the high-precision recognition requirement under a complex dynamic environment, but the key technical problem is how to maintain the stability and accuracy of the license plate recognition performance under the interference of real-time environment variables such as illumination intensity change, vehicle speed fluctuation and the like, and meanwhile, the system can be adaptively adjusted to deal with diversified scene challenges. The intense variation of illumination intensity (such as day and night alternation, shadow shielding or strong light direct irradiation) can cause unstable contrast of license plate characters in an image data stream, increase the risk of misrecognition or misrecognition, and the rapid fluctuation of vehicle speed (such as instantaneous acceleration or deceleration of a parking lot entrance) can cause image blurring or license plate region positioning deviation, and reduce the reliability of feature extraction. The dynamics of these environmental variables makes the conventional fixed threshold recognition algorithm difficult to adapt, resulting in frequent exceeding of the performance decay trend index by the preset threshold, affecting the robustness of the recognition module. In addition, the sensitivity of the sensor array may be deregulated due to environmental interference in long-term use, and the response curve is difficult to accurately reflect the actual scene requirement, so that the instability of algorithm threshold optimization is further aggravated. When the existing system processes complex scenes (such as dense parking lots, low-illumination underground garages or rain and fog weather), accurate modeling on the relevance of environment variables and recognition performance is lacking, and the threshold value set cannot be dynamically adjusted through a feedback loop mechanism, so that the recognition result and the intelligent parking scene are insufficient in applicability matching degree. Disclosure of Invention The invention aims to provide a recognition method for a curb plate, which aims to solve the problems that in the prior art, the dynamic performance of environmental variables makes a conventional recognition algorithm with a fixed threshold difficult to adapt, and performance attenuation trend indexes frequently exceed a preset threshold and influence the robustness of a recognition module. In order to solve the technical problems, the invention adopts the following technical scheme: A method of identifying a curb plate, the method comprising the steps of: S101, extracting a feature vector from environment variable data as an input basis of a trend model by collecting real-time environment variable data in a recognition process of a curb plate; S102, calculating a correlation coefficient between an environment variable and identification performance by adopting a trend model according to the extracted feature vector to obtain a performance attenuation trend index, wherein the performance attenuation trend index is used as a reference basis for subsequent parameter adjustment; S103, if the performance attenuation trend index exceeds a preset threshold, performing sensitivity calibration operation on the sensor array, and acquiring an adjusted sensor response curve as an intermediate result of dynamic optimization; S104, determining an updating range of an algorithm threshold from the adjusted sensor response curve, and fusing environment variable data through an iterative optimization method to obtain an optimized threshold set so as to adapt to the current dynamic environment; s105, after the optimized threshold value set is obtained, judging the matching degree of the threshold value set and the identification performance, if the matching degree is lower than the standard value, re-analyzing trend model output through a feedback loop mechanism, and determining further threshold fine tuning parameters; S106, updating intelligent system configuration by adopting the determined threshold fine tuning parameters to obtain an enhanced identification module for processing the image data stream of the license plate of the curb machine in a complex scene; s107, acquiring the processed image data stream from the enhanced recognition module, judging the applicability of the processed image data stream in the intelligent parking scene, and outputting a final license plate recognition result to support traffic management application if the applicability meets the requirement. According to the above technical solution, in S101, by collecting real-time environmental variable data in the recognition process of the guidepost locomotive brand, extracting feature vectors fro