CN107403162B - System and method for collecting and classifying vehicle notice number data - Google Patents
System and method for collecting and classifying vehicle notice number data Download PDFInfo
- Publication number
- CN107403162B CN107403162B CN201710639834.0A CN201710639834A CN107403162B CN 107403162 B CN107403162 B CN 107403162B CN 201710639834 A CN201710639834 A CN 201710639834A CN 107403162 B CN107403162 B CN 107403162B
- Authority
- CN
- China
- Prior art keywords
- classification
- data
- vehicle
- image data
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
- G06V20/54—Surveillance or monitoring of activities, e.g. for recognising suspicious objects of traffic, e.g. cars on the road, trains or boats
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/32—Normalisation of the pattern dimensions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/08—Detecting or categorising vehicles
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a system and a method for acquiring and classifying vehicle notice number data, which comprises an acquisition part and a classification part, wherein the classification part is connected with the acquisition part, and the acquisition part is used for constructing a vehicle notice number information list and is used for acquiring data to construct a database to be classified; the classification part is used for training data in the database according to the notice number information to obtain a classification model, reclassifying the data on the image information level, and iterating the training and classification processes to construct a vehicle image classification database. When the system is adopted, the vehicle image database is established based on the existing information in the vehicle management system, and reclassification is carried out according to the image information, so that the vehicle notice number data acquisition and classification system can quickly establish the vehicle image classification database for developing vehicle identification products, and manual data labeling is avoided.
Description
Technical Field
The invention relates to the field of intelligent transportation, in particular to the field of intelligent transportation vehicle identification and the rapid establishment of a related database, and particularly relates to a system and a method for acquiring and classifying vehicle bulletin number data.
Background
With the rapid development of deep learning on classification and detection problems, application scenes of the deep learning in the fields of transportation and public security are increasingly diversified, and the application range of the deep learning relates to vehicle brand recognition, license plate recognition, vehicle type recognition (cars, trucks and passenger cars), color recognition, face recognition and the like. The vehicle brand, model system and age identification function not only help to screen the fake-licensed vehicles, but also are gradually developing into important technical means in criminal investigation work, namely, suspected vehicles are searched and searched in massive video and image data according to the vehicle brand and model system, and the efficiency of the method is far higher than that of the traditional man-sea tactics.
However, in the actual product development process, the following three prominent problems are often inevitably required to be faced for realizing the brand, model and age identification functions of vehicles:
(1) the data demand is large. There are hundreds of common vehicle brands in the market, and there are thousands of vehicles of different models under all brands. It is well known that the number and quality of data sets have a direct impact on the classification training model, and the training data for each vehicle is hundreds of pieces to cover different situations under different viewing angles and light rays. Therefore, the total data amount needs to reach hundreds of thousands or even more.
(2) The data distribution is not uniform. In the past vehicle brand, money system and age identification work, the phenomenon of data distribution inclination acquired through a traffic gate or a monitoring video is found to be serious and mainly embodied in the following two aspects: firstly, in the same region, the probability distribution difference of different brand data is huge; and secondly, the distribution of the vehicle brand data is different between different areas.
(3) The model needs to be maintained and updated continuously. New vehicle models are still continuously published every year, and the identification models need to be updated regularly. The problem that needs to be solved urgently is solved by avoiding blind data acquisition and carrying out targeted acquisition and supplement on the lacking part of data.
The traditional data acquisition method has the advantages of high difficulty, low efficiency, high manual labeling cost and long time consumption. Based on the above analysis, there is a need for an easy to implement vehicle brand data collection and data processing scheme. Therefore, the vehicle notice number data acquisition and classification system provides a data acquisition method based on the vehicle notice number. The vehicle bulletin number refers to a name given by a manufacturer to vehicles having the same type, brand, kind, series, and body type. The content of the vehicle notice number code comprises five parts: enterprise name code, vehicle category code, main parameter code, product serial number, and enterprise self-defined code. The vehicle notice number is rich in content and unique. Collecting data by using the vehicle bulletin number as an index is more direct and efficient.
In addition, some vehicle notice number data have the same vehicle face or vehicle tail form structure, and can not be distinguished from traffic monitoring images or videos only. This is more common between different lines of the same brand. Whether in the model training or using stage, the bulletin numbers with the same car face or car tail need to be grouped into one type to improve the recognition rate and the user experience. Obviously, the efficiency is low, the speed is slow, and the reliability is not high when the same car face is found in thousands of car faces by a manual marking method. In order to solve the problem, the vehicle notice number data acquisition and classification system provides an automatic classification scheme for vehicle notice number data aiming at the characteristics of the vehicle notice number data, and realizes automatic and efficient combination of data with the same vehicle face or vehicle tail and different notice numbers.
Disclosure of Invention
The invention aims to overcome the defects in the prior art, provides a system and a method for acquiring and classifying vehicle notice number data, which are used for solving the problems that the data acquisition difficulty is high, the redundancy is high, the manual labeling efficiency is low, different notice numbers possibly have the same vehicle face or vehicle tail morphological structure and cannot be distinguished only from image or video information in the existing vehicle brand, model system and age identification product development process, and establishes a training database required for research and development for vehicle brand, model system and age classification and identification products.
In order to achieve the above object, the system and method for acquiring and classifying vehicle notice number data of the present invention specifically comprises the following steps:
the vehicle notice number data acquisition and classification system is mainly characterized by comprising an acquisition part and a classification part, wherein the classification part is connected with the acquisition part, and the acquisition part is used for constructing a vehicle notice number information list and is used for acquiring data to construct a database to be classified; the classification part is used for training data in the database according to the notice number information to obtain a classification model, reclassifying the data on the image information level, and iterating the training and classification processes to construct a vehicle image classification database. Preferably, the collecting part comprises:
the corresponding relation establishing module is used for establishing a vehicle bulletin number information list, and the vehicle bulletin number information list comprises vehicle bulletin numbers and vehicle basic information corresponding to the vehicle bulletin numbers;
the data acquisition module is used for acquiring data in the vehicle management system according to the vehicle notice numbers in the vehicle notice number information list and establishing a database matched with the vehicle notice number information list;
the classification part comprises:
the preprocessing module is used for distributing sample labels to the data in the database according to the classification information contained in the vehicle notice number, dividing the data in the database according to a preset rule and acquiring at least 3 data sets, and is also used for preprocessing the data in the database;
the classification model training module is used for training the data set according to the classification information to obtain a classification training model;
the classification model testing module is used for testing the classification training model obtained by the classification model training module according to the data set to obtain a testing result;
the test result counting module is used for counting the test results output by the classification model test module to obtain corresponding test indexes;
and the data merging module is used for judging whether the image information corresponding to the different vehicle notice number data is consistent according to the test index, and classifying and merging the data corresponding to the vehicle notice numbers with consistent image information.
More preferably, the basic information of the vehicle includes the brand, the money system and the chronological description information of the vehicle corresponding to the vehicle bulletin board.
Preferably, the database is an image database, and the image data in the image database includes image data corresponding to the vehicle notice number included in the vehicle notice number information list acquired by the data acquisition module.
Preferably, the image data includes head image data when the vehicle passes on the front side and tail image data when the vehicle passes off the back side.
Preferably, the preprocessing module is an image preprocessing module, the image preprocessing module allocates sample labels to the image data according to the classification information, and divides the image data into 3 mutually disjoint data sets according to a preset rule, which are respectively a training set, a verification set and a test set,
the training set is used for carrying out model training to obtain a classification training model;
the verification set is used for adjusting an unformed classification training model in the training process to enable the obtained classification training model to meet preset conditions;
the test set is used for testing the classification training model;
the preprocessing module preprocesses the data in the database, namely the image preprocessing module clips and scales the image data in the image database as required.
Preferably, the classification model training module performs deep learning training on the training set according to the sample labels of the image data in the training set to obtain a convolutional neural network as a classification training model, and the classification model training module further passes the image data in the verification set through an unformed classification training model to obtain test labels of the image data in the verification set, compares the test labels with the sample labels of the image data to obtain the recognition rate of the unformed classification training model on the image data in the verification set, and adjusts the relevant weight in the unformed classification training model according to the recognition rate until the recognition rate of the unformed classification training model on the image data in the verification set reaches a preset threshold.
Preferably, the classification model testing module inputs the image data in the test set into the classification training model, obtains a test label of the image data in the test set, and compares the test label with a sample label of the image data to obtain a comparison result, where the test result includes the test label and the comparison result.
Preferably, the test indexes in the test result statistic module include: the test indexes comprise the identification rate of the test set, the identification rate of the single type and the cross error rate among the types, wherein,
the test set identification rate is the ratio of the number of all correctly identified image data in the test set to the total number of the image data in the test set;
the single-class identification rate is the ratio of the number of correctly identified image data with certain class of sample labels in the test set by the classification training model to the total number of the image data with the class of sample labels in the test set;
the cross error rate is the probability that image data with a certain type of sample label is mistakenly identified as image data with another type of test label after passing through a classification training model.
Preferably, the system for acquiring and classifying vehicle notice number data further comprises a corresponding relationship updating module, and the corresponding relationship updating module is configured to update the vehicle notice number information table corresponding to the image data after the classification and merging processing.
The method for acquiring and classifying the vehicle notice number data by the vehicle notice number data acquisition and classification system is mainly characterized by comprising the following steps of:
(1) the acquisition part constructs a vehicle notice number information list, acquires data and constructs a database;
(2) and training and reclassifying the data in the database through the classification part to obtain a classification database.
Preferably, the step (1) is implemented by the acquisition part, and the step (2) is implemented by the classification part.
More preferably, the step (1) comprises the following steps:
(1.1) the corresponding relation establishing module establishes a vehicle bulletin number information list, and the vehicle bulletin numbers are matched and corresponding to the corresponding basic vehicle information one by one, wherein the basic vehicle information comprises the brand, the money system, the year description information and the license plate number of the vehicle corresponding to the vehicle bulletin number;
and (1.2) the data acquisition module acquires image data corresponding to the license plate number according to the license plate number corresponding to the vehicle notice number contained in the vehicle notice number information list.
Particularly preferably, the step (2) comprises the following steps:
(2.1) the preprocessing module preprocessing the image data in the image database;
(2.2) the classification model training module trains according to the data set to obtain a classification training model;
(2.3) the classification model testing module obtains a classification training model obtained by training of the classification model training module, tests the classification training model according to the data set and outputs a test result;
(2.4) the test result statistic module counts the test results to obtain test indexes;
and (2.5) the data merging module merges the data corresponding to the vehicle bulletin number according to the test index and the preset condition.
Preferably, the step (2.1) comprises the following steps:
(2.1.1) the preprocessing module assigning a sample label to the image data in the image database;
(2.1.2) dividing the image data in the image database into 3 mutually disjoint data sets according to a preset rule, wherein the 3 data sets are respectively a training set, a verification set and a test set.
Preferably, when the image data in the data set have different sizes, the image preprocessing module preprocesses the image data in the image database to clip and scale the image data in the data set to a preset size.
Preferably, the step (2.2) is:
the classification model training module carries out deep learning on the training set according to the sample labels of the image data in the training set to obtain a convolutional neural network as a classification training model, and in the deep learning process, the classification model training module passes the image data in the verification set through an unformed classification training model to obtain an output test label, compares the test label with the sample labels of the image data to obtain the recognition rate of the unformed classification training model on the image data in the verification set, and adjusts the relevant weight in the unformed classification training model according to the recognition rate until the recognition rate of the unformed classification training model on the image data in the verification set reaches a preset threshold value.
Preferably, the step (2.3) is:
the classification model testing module inputs the image data in the testing set into the classification training model, obtains a testing label of the image data in the testing set, compares the testing label with a sample label of the image data, and obtains a comparison result, wherein the testing result comprises the testing label and the comparison result.
Preferably, the step (2.5) is followed by the following steps:
(2.6) repeating the steps (2.1) to (2.6) until the test index obtained by the test result counting module meets the preset iteration exit condition, and the data merging module also completes the merging meeting the preset condition;
and (2.7) the corresponding relation updating module updates and replaces the vehicle notice number information list according to the data merging relation given in the step (2.6).
By adopting the system and the method for acquiring and classifying the vehicle notice number data, the data acquisition and classification efficiency is improved, the data quality is improved, and the labeling cost is reduced aiming at the vehicle notice number data acquisition and classification, so that the problems of blind data acquisition and high redundancy rate caused by directly acquiring a large amount of image data in the prior art can be well solved, and the problems of low data processing efficiency and reliability, high cost of manpower and time caused by the adoption of a manual labeling method for classifying and screening the data in the prior art are also solved. According to the method, vehicle bulletin numbers are adopted as index acquisition data according to the characteristics of vehicle data, the method is more direct and efficient, meanwhile, a deep learning algorithm classification training model is used for judging the characteristic similarity between the head data or the tail data of different bulletin numbers in a database and merging the head data or the tail data and the corresponding vehicle information, and compared with the manual labeling method in the prior art, the efficiency and the reliability are higher.
Drawings
FIG. 1 is a system block diagram of a vehicle bulletin number data acquisition and classification system.
Fig. 2 is a structural diagram of a vehicle bulletin number and a corresponding description information list.
Fig. 3 is a flow chart of the system for acquiring and classifying the vehicle notice number data according to the present invention.
Fig. 4 is a flowchart of the test result statistical module in the vehicle bulletin number data collection and classification system for performing test result statistics according to the present invention.
Fig. 5 is a flowchart of the system for collecting and classifying vehicle notice data according to the present invention, when the notice merging module merges the notice.
Detailed Description
In order to more clearly describe the technical contents of the present invention, the following further description is given in conjunction with specific embodiments.
In order to make the technical means, the creation characteristics, the achievement purposes and the effects of the invention easy to understand, the invention is further explained below by combining the specific drawings.
Referring to fig. 1, a system block diagram of a vehicle bulletin number data collection and classification system is shown. The vehicle notice number data acquisition and classification system efficiently acquires image data of the vehicle head and the vehicle tail according to the vehicle notice number as index information, and merges the notice number data with consistent morphological structures of the vehicle head and the vehicle tail, so as to provide a required data basis for realizing vehicle brands, money systems and year classification products. The vehicle notice number data acquisition and classification system comprises an acquisition part and a classification part, wherein the classification part is connected with the acquisition part, and the acquisition part is used for constructing a vehicle notice number information list and is used for acquiring data to construct a database to be classified; the classification part is used for training data in the database according to the notice number information to obtain a classification model, reclassifying the data on the image information level, and iterating the training and classification processes to construct a vehicle image classification database.
As shown in fig. 1, the system for acquiring and classifying vehicle notice number data mainly includes a correspondence establishing module, a data acquiring module, an image preprocessing module, a classification model training module, a classification model testing module, a testing result counting module, and a data merging module. The image preprocessing module, the classification model training module, the classification model testing module, the test result counting module and the data merging module belong to the classification part.
The corresponding relation establishing module is used for establishing a vehicle bulletin number information list, and the vehicle bulletin number information list comprises vehicle bulletin numbers and vehicle basic information corresponding to the vehicle bulletin numbers. In a specific embodiment, the correspondence establishing module establishes a vehicle announcement number information list through a vehicle management station or counting the existing vehicle announcement numbers in the market, and establishes the correspondence between each announcement number and the vehicle basic information. Referring to fig. 2, the vehicle bulletin number information list created by the correspondence relationship establishing module includes two columns, which are vehicle bulletin numbers and corresponding vehicle basic information, respectively, wherein the left column is information of all vehicle bulletin numbers in the current market; each row in the right column is vehicle basic information of a corresponding vehicle bulletin number, the vehicle basic information including: the information of the brand to which the vehicle belongs, the information of the money system under the brand to which the vehicle belongs, and the information of the age under the money system under which the vehicle belongs.
The data acquisition module is used for acquiring data in the vehicle management system according to the vehicle notice numbers in the vehicle notice number information list and establishing a database matched with the vehicle notice number information list. In a specific embodiment, the data acquisition module acquires the image data according to the bulletin number according to the vehicle bulletin number information list. The image data includes head image data and tail image data of the vehicle. In the specific embodiment, the data acquisition module acquires an image related to the vehicle position in the original image through a target detection algorithm, performs license plate recognition on the image of the vehicle position, and stores a vehicle screenshot matched with the notice number into an image database according to a license plate recognition result.
The preprocessing module is used for distributing sample labels to the data in the database according to the classification information contained in the vehicle notice number, dividing the data in the database according to a preset rule and obtaining at least 3 data sets, and is also used for preprocessing the data in the database. The preprocessing module is an image preprocessing module, the image preprocessing module allocates sample labels to the image data according to classification information, and divides the image data into 3 mutually disjoint data sets according to a preset rule, wherein the data sets are respectively a training set, a verification set and a test set, and the training set is used for carrying out model training to obtain a classification training model; the verification set is used for adjusting an unformed classification training model in the training process to enable the obtained classification training model to meet preset conditions; the test set is used for testing the classification training model; the preprocessing module preprocesses the data in the database, namely the image preprocessing module clips and scales the image data in the image database as required. In a specific embodiment, the image preprocessing module gives each sample label according to the classification information of the vehicle notice number, and divides the database into a training set, a verification set and a test set according to the proportion, so that the sizes of the sample images are uniform to the width and the height of pixels required by training.
The classification model training module is used for training the data set according to the classification information to obtain a classification training model, and in a specific embodiment, the classification model training module trains the classification model by using a training set and a verification set which are divided in the image preprocessing module and by adopting a deep learning algorithm according to a sample label.
The classification model testing module is used for testing the classification training model obtained by the classification model training module according to the data set to obtain a testing result. In a specific embodiment, the classification model testing module performs classification prediction on a test set divided in the image preprocessing module, assigns labels to data in the test set, outputs test labels of the data in the test set, and compares the test labels with sample labels assigned to the data in the test set to obtain a comparison result. The test results include a test label, a sample label, and a comparison result for the data.
The test result statistic module is used for counting the test results output by the classification model test module to obtain corresponding test indexes. In one embodiment, the test metrics include: the method comprises the steps of testing set identification rate, single-class identification rate and inter-class cross error rate, wherein the testing indexes comprise the testing set identification rate, the single-class identification rate and the inter-class cross error rate, and the testing set identification rate is the ratio of the number of all correctly identified image data in the testing set to the total number of the image data in the testing set; the single-class identification rate is the ratio of the number of correctly identified image data with certain class of sample labels in the test set by the classification training model to the total number of the image data with the class of sample labels in the test set; the cross error rate is the probability that image data with a certain type of sample label is mistakenly identified as image data with another type of test label after passing through a classification training model.
The data merging module is used for judging whether the image information corresponding to the different vehicle notice number data is consistent according to the test index, and classifying and merging the vehicle notice number corresponding data with consistent image information. In one embodiment, the data merging module analyzes each statistical result in the test result statistical module, classifies and merges the data meeting the merging condition and the corresponding bulletin number, and sorts and merges the relationship list according to the connectability between the merging results.
In one embodiment, the basic information of the vehicle includes the brand, the money system and the chronological description information of the vehicle corresponding to the vehicle bulletin number.
In a specific embodiment, the database is an image database, the image data in the image database includes image data corresponding to the vehicle notice number included in the vehicle notice number information list acquired by the data acquisition module, and the image data includes image data of a vehicle head when the front of the vehicle passes through and image data of a vehicle tail driven away from the back.
In a specific embodiment, the classification model training module performs deep learning training on the training set according to sample labels of image data in the training set to obtain a convolutional neural network as a classification training model, and the classification model training module further passes through an unformed classification training model on the image data in the verification set to obtain test labels of the image data in the verification set, compares the test labels with the sample labels of the image data to obtain the recognition rate of the unformed classification training model on the image data in the verification set, and adjusts the relevant weight in the unformed classification training model according to the recognition rate until the recognition rate of the unformed classification training model on the image data in the verification set reaches a preset threshold.
In a specific embodiment, the classification model testing module inputs the image data in the test set into the classification training model, obtains a test label of the image data in the test set, and compares the test label with a sample label of the image data to obtain a comparison result, where the test result includes the test label and the comparison result.
The vehicle notice number data acquisition and classification system is also provided with a corresponding relation updating module, and the corresponding relation updating module is used for updating a vehicle notice number information table corresponding to the image data after classification and merging processing.
In an embodiment, the correspondence updating module regenerates the vehicle brand, model system and year description information of the merged class according to the merged relationship list output by the notice number merging module, and updates the correspondence between the vehicle brand, model system and year description information and each notice number in the merged class.
The method for realizing the acquisition and classification of the vehicle notice number data by the vehicle notice number data acquisition and classification system comprises the following steps:
(1) the acquisition part constructs a vehicle notice number information list, acquires data and constructs a database;
(1.1) the corresponding relation establishing module establishes a vehicle bulletin number information list, and the vehicle bulletin numbers are matched and corresponding to the corresponding basic vehicle information one by one, wherein the basic vehicle information comprises the brand, the money system, the year description information and the license plate number of the vehicle corresponding to the vehicle bulletin number;
and (1.2) the data acquisition module acquires image data corresponding to the license plate number according to the license plate number corresponding to the vehicle notice number contained in the vehicle notice number information list.
(2) Training the data in the database through the training part to obtain a classification training model;
(2.1) the preprocessing module preprocessing the image data in the image database;
(2.1.1) the preprocessing module assigning a sample label to the image data in the image database;
(2.1.2) dividing the image data in the image database into 3 mutually disjoint data sets according to a preset rule, wherein the data sets are respectively a training set, a verification set and a test set;
(2.2) the classification model training module trains according to the data set to obtain a classification training model;
(2.3) the classification model testing module obtains a classification training model obtained by training of the classification model training module, tests the classification training model according to the data set and outputs a test result;
(2.4) the test result statistic module counts the test results to obtain test indexes;
(2.5) the notice number merging module merges vehicle notice numbers according to the test indexes and preset conditions;
(2.6) repeating the steps (2.1) to (2.6) until the test index obtained by the test result counting module meets the preset iteration exit condition, and the data merging module also completes the merging meeting the preset condition;
and (2.7) the corresponding relation updating module updates and replaces the vehicle notice number information list according to the data merging relation given in the step (2.6).
The step (1) is realized by the acquisition part, and the step (2) is realized by the classification part.
In a preferred embodiment, when the image data in the data set have different sizes, the image preprocessing module preprocesses the image data in the image database, so that the image data in the data set is cropped and scaled to a preset size.
In a specific embodiment, the classification model training module performs deep learning on the training set according to sample labels of image data in the training set to obtain a convolutional neural network as a classification training model, and in the deep learning process, the classification model training module obtains an output test label from the image data in the verification set through an unformed classification training model, compares the test label with the sample labels of the image data to obtain a recognition rate of the unformed classification training model on the image data in the verification set, and adjusts a relevant weight in the unformed classification training model according to the recognition rate until the recognition rate of the unformed classification training model on the image data in the verification set reaches a preset threshold.
In a specific embodiment, the classification model testing module inputs the image data in the test set into the classification training model, obtains a test label of the image data in the test set, and compares the test label with a sample label of the image data to obtain a comparison result, where the test result includes the test label and the comparison result.
Referring to fig. 3, in one embodiment, the method for acquiring and classifying the vehicle notice data according to the vehicle notice data acquisition and classification system includes the following steps:
(1) and establishing a vehicle brand, a money system and a year description information table which are in one-to-one correspondence with the vehicle notice numbers. In the system initialization process, the notice numbers of the head and the tail of the vehicle are the same as the corresponding relation table of the brand, the money system and the year description information of the vehicle.
(2) Before data acquisition is carried out by the data acquisition module, license plate numbers of vehicles corresponding to each vehicle bulletin number are acquired in vehicle management systems of various regions by taking the vehicle bulletin numbers as indexes, and the number of the license plate numbers corresponding to each bulletin number is not less than 10. And searching the front image of the vehicle in the traffic checkpoint systems of all regions according to the license plate information, and searching the back image of the vehicle in the electric warning system.
In a preferred embodiment, the number of images of the head or tail of the vehicle under each bulletin is no less than 400. And the target detection algorithm in the data acquisition module is a DPM (deformable Parts model) algorithm, the position areas of all vehicles in the original acquired image are intercepted through the DPM algorithm, license plate recognition is carried out on all intercepted vehicles, the recognition result is consistent with the original retrieval used license plate number, namely, the vehicle picture corresponding to the notice number is considered and stored in a database, an image database is constructed, and the image database stores the head image data or the tail image data of the vehicles which are in one-to-one correspondence with the notice numbers of the vehicles. And the vehicle screenshot with the recognition result inconsistent with the retrieved license plate number is not saved.
(3) When the image preprocessing module carries out sample label distribution on image data of an image database, sample label distribution is carried out according to notice number information, N notice numbers are N types of sample labels, and vehicles corresponding to each notice number are regarded as the same type and are distributed with the same sample label information. And in the subsequent iteration process, reclassifying the samples according to the merging relation list, and taking the merged notice number samples as one class and distributing sample label information.
And the image preprocessing module is used for dividing the data of the vehicle head or the vehicle tail into a training set, a verification set and a test set which are not intersected with each other according to the proportion of 50%, 20% and 30% of each type, and the image sizes in the data sets are uniformly cut and scaled to the pixel width and height required by training for the training module to use.
(4) And training the data in the training set by adopting a deep learning method, and training an AlexNet convolutional neural network as a classification training model to realize classification and identification of the input data. And when the recognition rate obtained by inputting the data in the verification set into the unformed classification training model tends to be convergent, the training of the classification training model is finished.
(5) And the classification model testing module tests the classification training model trained in the step 4 by using the test set, acquires a test label output after the data in the test set is input into the classification training model, compares the test label with a sample label of the data, and stores an original label, a prediction label and a comparison result of the data as a test result.
(6) Obtaining each item recognition rate of the test set through the test statistic module, referring to the figure4. Wherein the recognition rate R of the test set is the ratio of the number of samples in the test set that are correctly recognized to the total number of samples in the test set. Counting the recognition rate r of each classn(N is more than or equal to 1 and less than or equal to N), namely the ratio of the number of correctly identified samples in all the test samples of the current nth class (N is more than or equal to 1 and less than or equal to N) to the total number of the class samples. Counting the probability e of each class being mistakenly identified as other classesnm(N is more than or equal to 1 and less than or equal to N, m is more than or equal to 1 and less than or equal to N, and m is not equal to N), namely the ratio of the number of samples identified as the m-th class (m is more than or equal to 1 and less than or equal to N and m is not equal to N) in the current test samples of the N-th class (N is more than or equal to 1 and less than or equal to N) to the total number of samples of the. And saving and outputting the statistical result in a table form.
(7) Judging the merging relationship of the notice numbers according to the merging conditions, and the concrete process is as follows:
(7.1) judging whether the recognition rate R of the test set is greater than a set first threshold value, if so, performing no merging operation on the current data set, otherwise, performing the merging operation, and entering the step (7.2);
(7.2) referring to fig. 5, the specific steps of the announcement merging module are as follows:
(7.2.1) judging the recognition rate r of the current class n one by one according to the classnIf the current class is not greater than the set second threshold, the current class does not need to be merged with other classes if the current class meets the condition, and if the current class does not meet the condition, the next step of judgment is carried out;
(7.2.2) judging the probability e of the current class n being mistakenly identified as the m-th class one by onenmRecognition rate r with class nnWhether the ratio of the first to the second classes is larger than a set third threshold value or not, if the ratio of the first to the second classes is not larger than the set third threshold value, the nth class and the mth class are similar enough and need not to be combined;
and (7.3) arranging the merged relation list according to the connectability among the merged relations, wherein the subclasses of each large class after merging have the same vehicle face or vehicle tail shape structure. And redistributing the labels corresponding to the merged bulletin numbers. Publication numbers that are grouped together into a class have the same label, and in a preferred embodiment, the label value is the minimum of the labels in its subclass. The classes that are not merged remain unchanged from the original tags.
(8) Judging whether the output result of the step 7 contains classes to be merged, if so, repeating the steps 3 to 7 untilThe recognition rate R of the classification model on the test set meets the preset iteration exit condition R > tR(tRA fourth threshold), or there is no class meeting the merging condition, the data merging is completed. The fourth threshold value t during actual useR＝0.9。
(9) And (3) updating the corresponding relation table of the notice numbers of the vehicle head or the vehicle tail and the brand, the money system and the year description information of the vehicle built in the step (1) according to the merging result of the notice numbers, wherein the modification process is as follows:
(9.1) judging whether each merged class consists of a plurality of bulletin number subclasses one by one, if only one bulletin number subclass exists in the current class, not modifying corresponding vehicle brand, money system and year description information, and if not, entering the next step;
(9.2) extracting the vehicle brand, money system and age information corresponding to all the notice numbers in the current class, such as: if the current class is formed by merging k bulletin numbers, the description information corresponding to the bulletin numbers is as follows: brand xx-series xx-chronous xx1, brand xx-series xx-chronous xx2 … … brand xx-series xx-chronous xxk;
(9.3) splicing the description information of each bulletin number to generate new description information of the current class, for example: brand xx-series xx-chronous xx1 or brand xx-series xx-chronous xx2 … … or brand xx-series xx-chronous xxk;
and modifying the basic description information of the vehicle corresponding to each vehicle notice number in the current class into the new description information generated in the step (9.3) in the corresponding relation table consisting of the vehicle notice numbers and the basic information of the vehicle. And updating the corresponding relation table of the vehicle head or the vehicle tail notice number, the vehicle brand, the vehicle system and the year description information.
By adopting the system and the method for acquiring and classifying the vehicle notice number data, the data acquisition and classification efficiency is improved, the data quality is improved, and the labeling cost is reduced aiming at the vehicle notice number data acquisition and classification, so that the problems of blind data acquisition and high redundancy rate caused by directly acquiring a large amount of image data in the prior art can be well solved, and the problems of low data processing efficiency and reliability, high cost of manpower and time caused by the adoption of a manual labeling method for classifying and screening the data in the prior art are also solved. According to the method, vehicle bulletin numbers are adopted as index acquisition data according to the characteristics of vehicle data, the method is more direct and efficient, meanwhile, a deep learning algorithm classification training model is used for judging the similarity between the data of the vehicle heads or the vehicle tails with different bulletin numbers and carrying out data combination, and compared with a manual labeling method in the prior art, the method is higher in efficiency and reliability.
In this specification, the invention has been described with reference to specific embodiments thereof. It will, however, be evident that various modifications and changes may be made thereto without departing from the broader spirit and scope of the invention. The specification and drawings are, accordingly, to be regarded in an illustrative rather than a restrictive sense.
Claims (8)
1. A vehicle notice number data acquisition and classification system is characterized by comprising an acquisition part and a classification part, wherein the classification part is connected with the acquisition part, and the acquisition part is used for constructing a vehicle notice number information list and is used for acquiring data to construct a database to be classified; the classification part is used for training data in the database according to the notice number information to obtain a classification model, reclassifying the data on the image information level, and iterating the training and classification processes to construct a vehicle image classification database;
the acquisition part comprises:
the corresponding relation establishing module is used for establishing a vehicle bulletin number information list, and the vehicle bulletin number information list comprises vehicle bulletin numbers and vehicle basic information corresponding to the vehicle bulletin numbers;
the data acquisition module is used for acquiring data in the vehicle management system according to the vehicle notice numbers in the vehicle notice number information list and establishing a database matched with the vehicle notice number information list;
the classification part comprises:
the preprocessing module is used for distributing sample labels to the data in the database according to the classification information contained in the vehicle notice number, dividing the data in the database according to a preset rule and acquiring at least 3 data sets, and is also used for preprocessing the data in the database;
the classification model training module is used for training the data set according to the classification information to obtain a classification training model;
the classification model testing module is used for testing the classification training model obtained by the classification model training module according to the data set to obtain a testing result;
the test result counting module is used for counting the test results output by the classification model test module to obtain corresponding test indexes;
and the data merging module is used for judging whether the image information corresponding to the different vehicle notice number data is consistent according to the test index, and classifying and merging the data corresponding to the vehicle notice numbers with consistent image information.
2. The system according to claim 1, wherein the basic information of the vehicle includes brand, model system and age description information of the vehicle corresponding to the vehicle bulletin number;
the database is an image database, and image data in the image database comprises image data corresponding to the vehicle notice numbers contained in the vehicle notice number information list acquired by the data acquisition module;
the image data comprises head image data when the front surface of the vehicle passes through and tail image data when the back surface of the vehicle drives away;
the preprocessing module is an image preprocessing module, the image preprocessing module allocates sample labels to the image data according to classification information, and divides the image data into 3 mutually disjoint data sets according to a preset rule, wherein the data sets are respectively a training set, a verification set and a test set,
the training set is used for carrying out model training to obtain a classification training model;
the verification set is used for adjusting an unformed classification training model in the training process to enable the obtained classification training model to meet preset conditions;
the test set is used for testing the classification training model;
the preprocessing module preprocesses the data in the database, namely the image preprocessing module clips and scales the image data in the image database as required.
3. The vehicle bulletin number data collecting and classifying system according to claim 2, the classification model training module is used for training the classification model according to the sample labels of the image data in the training set, deep learning training is carried out on the training set to obtain a convolutional neural network as a classification training model, and the classification model training module also obtains the test label of the image data in the verification set through the unformed classification training model, comparing the test label with the sample label of the image data to obtain the recognition rate of the unformed classification training model to the image data in the verification set, adjusting the relevant weight in the unformed classification training model according to the recognition rate until the recognition rate of the unformed classification training model to the image data in the verification set reaches a preset threshold value;
the classification model testing module inputs the image data in the testing set into the classification training model, obtains a testing label of the image data in the testing set, compares the testing label with a sample label of the image data, and obtains a comparison result, wherein the testing result comprises the testing label and the comparison result;
the test indexes in the test result statistic module comprise: test set identification rate, single class identification rate, and inter-class cross-over error rate, wherein,
the test set identification rate is the ratio of the number of all correctly identified image data in the test set to the total number of the image data in the test set;
the single-class identification rate is the ratio of the number of correctly identified image data with certain class of sample labels in the test set by the classification training model to the total number of the image data with the class of sample labels in the test set;
the cross error rate is the probability that image data with a certain type of sample label is mistakenly identified as image data with another type of test label after passing through a classification training model.
4. The system according to claim 1, further comprising a correspondence update module, wherein the correspondence update module is configured to update the vehicle notice number information table corresponding to the classified and combined image data.
5. A method for realizing the acquisition and classification of vehicle notice number data based on the system of any one of claims 1 to 4, characterized in that the method comprises the following steps:
(1) the acquisition part constructs a vehicle notice number information list, acquires data and constructs a database;
(2) training and reclassifying the data in the database through the classification part to obtain a classification database;
the step (1) is realized by the acquisition part, wherein the tags correspond to vehicle bulletin numbers one to one, and the step (1) comprises the following steps:
(1.1) the corresponding relation establishing module establishes a vehicle bulletin number information list, and the vehicle bulletin numbers are matched and corresponding to the corresponding basic vehicle information one by one, wherein the basic vehicle information comprises the brand, the money system, the year description information and the license plate number of the vehicle corresponding to the vehicle bulletin number;
(1.2) the data acquisition module acquires image data corresponding to the license plate number according to the license plate number corresponding to the vehicle notice number contained in the vehicle notice number information list;
the step (2) is realized by a classification part, and the step (2) comprises the following steps:
(2.1) the preprocessing module preprocessing the image data in the image database;
(2.2) the classification model training module trains according to the data set to obtain a classification training model;
(2.3) the classification model testing module obtains a classification training model obtained by training of the classification model training module, tests the classification training model according to the data set and outputs a test result;
(2.4) the test result statistic module counts the test results to obtain test indexes;
and (2.5) the data merging module merges the data corresponding to the vehicle bulletin number according to the test index and the preset condition.
6. The method of claim 5 for enabling vehicle bulletin number data collection and classification,
the preprocessing module is an image preprocessing module, the image preprocessing module allocates sample labels to the image data according to classification information, and divides the image data into 3 mutually disjoint data sets according to a preset rule, wherein the data sets are respectively a training set, a verification set and a test set,
the training set is used for carrying out model training to obtain a classification training model;
the verification set is used for adjusting an unformed classification training model in the training process to enable the obtained classification training model to meet preset conditions;
the test set is used for testing the classification training model;
the preprocessing module preprocesses the data in the database to enable the image preprocessing module to crop and scale the image data in the image database as required,
the step (2.1) comprises the following steps:
(2.1.1) the preprocessing module assigning a sample label to the image data in the image database;
(2.1.2) dividing the image data in the image database into 3 mutually disjoint data sets according to a preset rule, wherein the data sets are respectively a training set, a verification set and a test set;
and when the image data in the data set have different sizes, the image preprocessing module preprocesses the image data in the image database to cut and scale the image data of the data set to a preset size.
7. The method for realizing the acquisition and classification of the vehicle bulletin data according to the claim 6, wherein the step (2.2) is as follows:
the classification model training module carries out deep learning on the training set according to the sample labels of the image data in the training set to obtain a convolutional neural network as a classification training model, and in the deep learning process, the classification model training module passes the image data in the verification set through an unformed classification training model to obtain an output test label, compares the test label with the sample labels of the image data to obtain the recognition rate of the unformed classification training model on the image data in the verification set, and adjusts the relevant weight in the unformed classification training model according to the recognition rate until the recognition rate of the unformed classification training model on the image data in the verification set reaches a preset threshold value;
the step (2.3) is as follows:
the classification model testing module inputs the image data in the testing set into the classification training model, obtains a testing label of the image data in the testing set, compares the testing label with a sample label of the image data, and obtains a comparison result, wherein the testing result comprises the testing label and the comparison result.
8. The method for acquiring and classifying vehicle bulletin data according to claim 5, wherein the vehicle bulletin data acquiring and classifying system further comprises:
the corresponding relation updating module is used for updating the vehicle notice number information table corresponding to the classified and combined image data;
the step (2.5) is followed by the following steps:
(2.6) repeating the steps (2.1) to (2.6) until the test index obtained by the test result counting module meets the preset iteration exit condition, and the data merging module also completes the merging meeting the preset condition;
(2.7) the corresponding relation updating module updates and replaces the vehicle notice number information list according to the data merging relation given in the step (2.6);
the test indexes comprise various indexes of a test set identification rate, a single-class identification rate and an inter-class cross error rate, wherein,
the test set identification rate is the ratio of the number of all correctly identified image data in the test set to the total number of the image data in the test set;
the single-class identification rate is the ratio of the number of correctly identified image data with certain class of sample labels in the test set by the classification training model to the total number of the image data with the class of sample labels in the test set;
the cross error rate is the probability that image data with a certain type of sample label is mistakenly identified as image data with another type of label after passing through a classification training model.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710450924 | 2017-06-15 | ||
| CN2017104509245 | 2017-06-15 |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107403162A CN107403162A (en) | 2017-11-28 |
| CN107403162B true CN107403162B (en) | 2020-11-03 |
Family
ID=60402480
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710639834.0A Active CN107403162B (en) | 2017-06-15 | 2017-07-31 | System and method for collecting and classifying vehicle notice number data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107403162B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110786847B (en) * | 2018-08-02 | 2022-11-04 | 深圳市理邦精密仪器股份有限公司 | Electrocardiogram signal library building method and analysis method |
| US10818042B1 (en) | 2020-01-14 | 2020-10-27 | Capital One Services, Llc | Vehicle information photo overlay |
| US10832400B1 (en) | 2020-01-14 | 2020-11-10 | Capital One Services, Llc | Vehicle listing image detection and alert system |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2013148971A (en) * | 2012-01-17 | 2013-08-01 | Oki Electric Ind Co Ltd | Vehicle type discrimination device |
| CN103927877A (en) * | 2014-04-01 | 2014-07-16 | 内蒙古银安科技开发有限责任公司 | Motor vehicle image information storage reading method and system |
| CN106022285A (en) * | 2016-05-30 | 2016-10-12 | 北京智芯原动科技有限公司 | Vehicle type identification method and vehicle type identification device based on convolutional neural network |
| CN106295541A (en) * | 2016-08-03 | 2017-01-04 | 乐视控股（北京）有限公司 | Vehicle type recognition method and system |
| CN106446949A (en) * | 2016-09-26 | 2017-02-22 | 成都通甲优博科技有限责任公司 | Vehicle model identification method and apparatus |
-
2017
- 2017-07-31 CN CN201710639834.0A patent/CN107403162B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2013148971A (en) * | 2012-01-17 | 2013-08-01 | Oki Electric Ind Co Ltd | Vehicle type discrimination device |
| CN103927877A (en) * | 2014-04-01 | 2014-07-16 | 内蒙古银安科技开发有限责任公司 | Motor vehicle image information storage reading method and system |
| CN106022285A (en) * | 2016-05-30 | 2016-10-12 | 北京智芯原动科技有限公司 | Vehicle type identification method and vehicle type identification device based on convolutional neural network |
| CN106295541A (en) * | 2016-08-03 | 2017-01-04 | 乐视控股（北京）有限公司 | Vehicle type recognition method and system |
| CN106446949A (en) * | 2016-09-26 | 2017-02-22 | 成都通甲优博科技有限责任公司 | Vehicle model identification method and apparatus |
Non-Patent Citations (2)
| Title |
|---|
| Advanced scale-space, invariant, low detailed feature recognition from images - car brand recognition;Stefan Badura et al.;《Proceedings of the International Multiconference on Computer Science and Information Technology》;20110106;第19-23页 * |
| 基于LLC与加权SPM的车辆品牌型号识别;李熙莹 等;《计算机工程》;20170515;第43卷(第5期);第210-216页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107403162A (en) | 2017-11-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| DE112013001858B4 (en) | Multiple-hint object recognition and analysis | |
| CN112016605B (en) | Target detection method based on corner alignment and boundary matching of bounding box | |
| CN107403162B (en) | System and method for collecting and classifying vehicle notice number data | |
| US11113582B2 (en) | Method and system for facilitating detection and identification of vehicle parts | |
| CN111814472B (en) | Text recognition method, device, equipment and storage medium | |
| Heydarian et al. | Automated visual recognition of construction equipment actions using spatio-temporal features and multiple binary support vector machines | |
| CN105117737A (en) | Method and apparatus for determining real direction of vehicle on basis of locus vector of vehicle | |
| Sikirić et al. | Classifying traffic scenes using the GIST image descriptor | |
| Sikirić et al. | Image representations on a budget: Traffic scene classification in a restricted bandwidth scenario | |
| CN105975949A (en) | Visual-information-based automobile identification method | |
| Cao et al. | An end-to-end neural network for multi-line license plate recognition | |
| CN107273889B (en) | License plate recognition method based on statistics | |
| Balali et al. | Video-based detection and classification of US traffic signs and mile markers using color candidate extraction and feature-based recognition | |
| Fernando et al. | Automatic road traffic signs detection and recognition using ‘You Only Look Once’version 4 (YOLOv4) | |
| Thomas et al. | Smart car parking system using convolutional neural network | |
| CN112508000B (en) | Method and equipment for generating OCR image recognition model training data | |
| Laroca et al. | Do we train on test data? The impact of near-duplicates on license plate recognition | |
| de Almeida et al. | Vehicle Occurrence-based Parking Space Detection | |
| CN111611933A (en) | Information extraction method and system for document image | |
| KR20200077257A (en) | Method and apparatus for automatically generating learning data for machine learning | |
| CN113076917B (en) | Pedestrian quality evaluation method and system | |
| CN114973169A (en) | Vehicle classification counting method and system based on multi-target detection and tracking | |
| US10311084B2 (en) | Method and system for constructing a classifier | |
| CN112380913A (en) | License plate detection and identification method based on combination of dynamic adjustment and local feature vector | |
| Ebrahimi et al. | Car license plate recognition using color features of Persian license plates |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |