CN109544707B

CN109544707B - Intelligent control system and method for access of parking lot

Info

Publication number: CN109544707B
Application number: CN201710840617.8A
Authority: CN
Inventors: 凌广明; 丁勇; 徐航; 张镇; 徐爱萍; 徐武平
Original assignee: Wuhan Wondnet Technology Co ltd; Wuhan University WHU
Current assignee: Wuhan Wondnet Technology Co ltd; Wuhan University WHU
Priority date: 2017-09-18
Filing date: 2017-09-18
Publication date: 2021-06-01
Anticipated expiration: 2037-09-18
Also published as: CN109544707A

Abstract

The invention discloses an intelligent control system and method for access of a parking lot, wherein the system comprises a camera I, a camera II, a camera III, a monitoring system and a charging system, the camera I and the camera II are arranged at an entrance of the parking lot, the camera III is arranged at an exit of the parking lot, the camera I, the camera II, the camera III and the charging system are all connected with the monitoring system, and the monitoring system comprises a heterogeneous dual-camera license plate recognition system based on confidence and a fuzzy matching system for exit license plate recognition. The invention can design a double-camera recognition system and an algorithm based on four cameras of different types and different models by using a multithreading and neural network technology, solves the problem that any lane of a monitoring system supports the combination of the cameras of different types and different models to form double-camera recognition, and solves the problem that a vehicle automatically leaves the field under the condition of wrong camera recognition.

Description

Intelligent control system and method for access of parking lot

Technical Field

The invention relates to the field of intelligent technical control, in particular to a system and a method for intelligently controlling access of a parking lot.

Background

Nowadays, with the development of economy and the continuous improvement of science and technology in China, the quantity of automobile reserves in China is rapidly increased, and the problem of difficult parking in many cities is gradually highlighted. How to solve the problem of difficult parking is urgent. Currently, many cities have parking lots as components of smart cities, and are gradually combined with other smart cities. It has become increasingly important to design an intelligent parking lot charging management system.

Current parking fee software is typically manufactured for cameras of a particular model of a company, and cameras manufactured by other companies cannot use the software, thereby reducing the efficiency of software development. In addition, due to the problem of camera identification angle, parking lot cameras in some areas cannot capture the license plate of a vehicle every time, and two cameras are required to be installed for identification. The system is required to support the joint monitoring of one lane using two cameras at the same time. General toll software on the market today does not support dual camera identification. In addition, some systems support dual-camera recognition, but when one camera is damaged, the camera can be replaced by the same type of camera, and the cameras of other types cannot be used. However, in the current mainstream cameras, the recognition accuracy rate is good or bad in different environments, so that a system is required to support that one lane is monitored by using two cameras of different types or different models; the problem of wrong recognition can exist due to the problem of correct recognition rate of the cameras, the situation that the license plate recognition of the cameras of different types and different models is wrong in different environments can be caused, and the automatic leaving and intelligent charging of the vehicle are difficult due to the wrong recognition result every time, so that the automatic leaving failure is caused. Therefore, to implement intelligent monitoring of the parking lot, customers are required to overcome the defects of the prior art.

Disclosure of Invention

The invention aims to overcome the defects in the prior art and provides an intelligent control system and method for access of a parking lot.

The technical scheme for realizing the purpose of the invention is as follows: the intelligent control system for the entrance and the exit of the parking lot comprises a first camera, a second camera, a third camera, a monitoring system and a charging system, wherein the first camera and the second camera are arranged at an entrance of the parking lot, the third camera is arranged at an exit of the parking lot, the first camera, the second camera, the third camera and the charging system are all connected with the monitoring system, and the monitoring system comprises a heterogeneous dual-camera license plate recognition system based on confidence coefficient and a fuzzy matching system for the license plate recognition of the exit.

As optimization, the intelligent control system further comprises a charging system connected with the monitoring system, and the charging system is an ETC system.

An intelligent control method for access in a parking lot comprises the following steps:

1) when a monitoring system of the intelligent control system for access of the parking lot monitors that the vehicle enters a lane, the monitoring system sends a command to a camera I and a camera II;

2) after receiving the command, the first camera and the second camera acquire vehicle information and transmit acquired data to a monitoring system;

3) a heterogeneous dual-camera license plate recognition system based on confidence in the monitoring system recognizes the data collected in the step 2);

4) when the vehicle exits from the parking lot, when the data identified in step 3) is correct, directly going to step 8); when the data identified in the step 3) is wrong, directly going to the step 5);

5) when the vehicle exits the parking lot, the monitoring system of the intelligent parking lot entrance and exit control system monitors that the vehicle exits the lane, and the monitoring system sends a command to the camera III;

6) after receiving the command, the camera III collects the vehicle information and transmits the collected data to the monitoring system;

7) a fuzzy matching system for recognizing the license plate at the departure in the monitoring system recognizes the data collected in the step 6) and matches the data collected in the step 3);

8) the monitoring system sends the correct identification data in the step 3) or the matched data result in the step 7) to a charging system for processing, and the charging system feeds back charging information to the monitoring system;

as an optimization, the algorithm of the confidence-based heterogeneous dual-camera license plate recognition system comprises the following steps:

1) acquiring camera parameters of corresponding lanes, initializing CI, and starting a camera identification thread T₁,T₂And waiting for triggering.

2) Triggering a vehicle arrival recognition thread, acquiring camera information, recording license plate information, namely a license plate as P, and converting into a first confidence coefficient Dg;

firstly, judging whether the CI is locked, if so, waiting, otherwise, turning to the second step;

judging CI.plate _ last as P according to CI, if TRUE, turning to 3), otherwise, turning to the third step;

calculating the time difference Span (CI. time-DateTime. now () according to CI), judging that Span is more than 250ms, if TRUE, turning to the fourth step, otherwise, turning to the fifth step;

the description is a new record, namely another vehicle, writes the confidence coefficient Dg, the license plate P and the recognition time into CI, CI.plate is P, CI.last _ plate is CI.plate, CI.Dg is Dg, and CI.time is DateTime.now (), and inserts the license plate information into the database, turn 3);

judging whether the camera is the same as the camera identified last time, if not, turning to nine, otherwise, turning to nine;

sixthly, judging whether the confidence coefficient of the two-phase machine is greater than a threshold value cr, if so, rotating 3), and if not, rotating seventhly;

seventhly, the confidence coefficient of comparison CI.Dg is larger than Dg, if FAST is found, go to 3), otherwise go to the method;

and (2) acquiring a license plate P and a confidence coefficient Dg, updating a structure CI, wherein the CI is equal to P, the CI is equal to CI, the last is equal to CI, and the CI is equal to Dg, and modifying license plate information in the database. Turn 3);

ninthly, judging whether the model is the same, if not, converting to (at the moment, the confidence of conversion and comparison is that the confidence of the camera is not an artificial confidence), otherwise, converting to (r);

the method comprises the following steps that (1) the weight value trained by a BP neural network and the CI are utilized to obtain the model of a camera which is accurately identified, if the model of the camera is the identified model of the camera, the model is changed to be phi, and if not, the model is changed to be 3);

3) the algorithm ends.

As an optimization, theThe Span is the time difference between the time when the camera recognizes the license plate at this time and the time when the camera recognizes the license plate at the last time, the algorithm of the heterogeneous dual-camera license plate recognition system based on the confidence coefficient is started by two threads, each camera corresponds to one thread, and the recognition thread started by the first camera is T₁The identification thread of the second camera is turned on is T₂The two threads share the same structural body CI respectively, and read and write are carried out on the CI when T is reached₁When reading from or writing to CI, T₂Wait until T₁After the reading and writing is finished, T₂When reading and writing the same CI, T₁Wait until T₂And finishing reading and writing.

As optimization, the algorithm flow of the fuzzy matching system for the outgoing license plate recognition is as follows:

1) acquiring a license plate P identified by a camera;

2) acquiring a lane similarity threshold value W, acquiring a license plate array [ T ] of a field vehicle in the database, wherein the size of the counting array is N, and setting i as 0; firstly, turning;

judging that i is equal to N, and i + +, if the i is FALSE, turning to the second, and if the i is TRUE, turning to the third;

judging that P is T [ i ], if the P is TRUE, turning to 3), otherwise, turning to the first step;

assigning i to be 0, and taking the on-site vehicle T [ i ] and P to carry out general fuzzy matching (i is changed from 0 to N) to be converted to (iv);

judging that i is equal to N, and if i is FALSE, turning to fifth step, if TRUE, turning to sixth step;

using similarity formula 1 to calculate the similarity between two license plates

If D ═ W, record the license plate T [ i ═ W]If not, directly transferring to the fourth step;

sixthly, judging whether there is similar license plate record, if it is TRUE, taking out only license plate T x]Calling dictionary-based self-learning algorithm to calculate similar character pairs, and turning to

Otherwise, turning to the seventh step;

assigning i to 0, starting to take the field vehicle T [ i ] and P to perform fuzzy matching (i ranges from 0 to N) based on the longest common subsequence, and turning to the step b;

judging if i is N and i + +, if FALSE, go to step ninc, if TRUE, go to step OnR;

ninthly, calculating the similarity of the two license plates by using a fuzzy matching formula based on the longest public subsequence, recording the license plate T [ i ] if D ═ W, and turning to the license plate T [ i ], or else, directly turning to the license plate T [ i ];

r (R) judges if there is similar license plate record, if it is TRUE, it takes out only license plate T [ x ]]Go to

Otherwise, turning to manual processing and turning to 3);

license plate T [ x ] with maximum similarity]Go to 3);

3) the algorithm ends.

In the algorithm of the fuzzy matching system for the outgoing license plate recognition, the similarity is D, P is the license plate recognized by the camera, T is the license plate to be matched, and the fuzzy matching threshold is W and x_iIs the ith character of P, y_iThe ith character of T, d is the similarity weight of the character pair and d ═ f (x)_iy_j) And L is the length of the license plate.

The invention has the positive effects that:

(1) aiming at the characteristics of different cameras, the invention can design a double-camera recognition system and an algorithm based on four cameras of different types and different models by using multithreading and neural network technology, and solves the problem that any lane of a monitoring system supports the combination of the cameras of different types and different models to form double-camera recognition.

(2) A similar dictionary based on self-learning, a fuzzy matching system and an algorithm are provided, so that the automatic departure of the vehicle can be realized to a great extent under the condition that the camera identifies the vehicle license plate wrongly. The experimental result shows that the fuzzy matching algorithm improves the intelligence degree of the parking lot system by about 3 percent, thereby solving the problem that the vehicle automatically leaves the field under the condition of wrong camera identification.

Drawings

FIG. 1 is a schematic structural diagram of an intelligent control system for entrance and exit of a parking lot according to the present invention;

FIG. 2 is a model diagram of a three-layer neural network in the confidence-based heterogeneous dual-camera license plate recognition system of the present invention;

FIG. 3 is a data diagram of a neural network test experiment in a confidence-based heterogeneous dual-camera license plate recognition system of the present invention;

FIG. 4 is a process diagram of neural network training in a confidence-based heterogeneous dual-camera license plate recognition system of the present invention;

FIG. 5 is an analysis diagram of neural network regression in a confidence-based heterogeneous dual-camera license plate recognition system of the present invention;

FIG. 6 is a block diagram of dual camera data in a confidence based heterogeneous dual camera license plate recognition system of the present invention;

FIG. 7 is a flow chart of a main thread starting recognition thread in the confidence-based heterogeneous dual-camera license plate recognition system of the present invention;

FIG. 8 is a flow chart of a recognition thread in the confidence-based heterogeneous dual-camera license plate recognition system of the present invention;

FIG. 9 is a similar dictionary graph used by the algorithm in the fuzzy matching system for license plate identification at the departure of the present invention;

FIG. 10 is a diagram of a self-learning algorithm for a similar dictionary used by the algorithm in the fuzzy matching system for license plate recognition at departure of the present invention;

FIG. 11 is a flow chart of a fuzzy matching algorithm in the fuzzy matching system for license plate identification at the departure of the present invention;

FIG. 12 is a graph of generic fuzzy matching test data;

FIG. 13 is a fuzzy matching test data chart in the fuzzy matching system for license plate identification at the departure of the present invention;

FIG. 14 is a communication parameter configuration diagram of the present invention;

FIG. 15 is a monitoring interface diagram of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

As shown in fig. 1, the present invention is an intelligent control system for entrance and exit of a parking lot, which is composed of a first camera, a second camera, a third camera, a monitoring system and a charging system, wherein the first camera and the second camera are arranged at an entrance of the parking lot, the third camera is arranged at an exit of the parking lot, the first camera, the second camera, the third camera and the charging system are all connected with the monitoring system, the monitoring system comprises a heterogeneous dual-camera license plate recognition system based on confidence and a fuzzy matching system for exit license plate recognition, the intelligent control system further comprises a charging system connected with the monitoring system, and the charging system is an ETC system.

in the invention, the algorithm of the heterogeneous dual-camera license plate recognition system based on confidence coefficient comprises the following contents and steps:

firstly, the confidence of camera recognition is the degree of judgment of the accuracy of the license plate recognized by the camera, and the camera returns a corresponding floating point number every time the license plate is recognized by the camera, which is called as the confidence of recognition, and is called as the confidence for short.

The confidence coefficient is a floating point number with a decimal number between (0 and 1), and the higher the confidence coefficient is, the more accurate the license plate recognized by the camera at this time is.

The confidence returned by the camera is based on the own functions of a certain type of camera, and the confidence of different types of cameras cannot be directly compared. For example: in the actual test, the result of identifying a certain license plate by the contact short 1 type camera is more accurate than the result of identifying the contact short 2 type camera, but the corresponding confidence coefficient is lower than that of the contact short 2 type camera, so the confidence coefficients can be directly compared only when the cameras are of the same type and the cameras of the same type identify the same license plate, and the confidence coefficients of the cameras of different types cannot be directly compared. For the situation, a similar pattern recognition method is adopted, and a neural network model is used for judging the recognition result of the selected camera.

As shown in the three-layer neural network model of fig. 2, the BP neural network is a multi-layer feedforward neural network, and the main characteristics of the network are signal forward transmission and error backward propagation. In the forward pass, the input signal goes through the processing of the hidden layer to the output layer. If the desired output is not available at the output layer, then we turn to back propagation. According toThe error between the output and the desired output adjusts the weight and threshold. The neural network model structure of this case is shown in fig. 2, which is a BP neural network topology structure with 4 input/output neurons. Wherein x₁,x₂,x₃,x₄Is the input value of the network, corresponding to the confidence of the four cameras, y₁,y₂,y₃,y₄The output values of the network are, one of the normal output values should be 1, and the other three should be 0, if the output is {0,1,0,0} which indicates that the recognition result of the second camera is more accurate, the BP neural network needs to be trained before use, and the training of the standard BP comprises the following steps:

the first step is as follows: initialization of the network. Determining the number of neurons of the input layer, the number of neurons of the output layer, and the connection weight w between the hidden layer and the output layer according to the input and the output to be trained_ij，w_jk. Initializing a threshold value a of the hidden layer and a threshold value b of the output layer. The learning rate and neuron excitation functions of the neural network are given.

The second step is that: computation of the hidden layer output. According to the weight w between the input layer, the input layer and the hidden layer_ijThe output H of the hidden layer is calculated as the threshold a of the hidden layer.

Where l is the number of nodes in the hidden layer, f is the excitation function, n is the number of neurons in the input layer, and n is 4 in this document.

The third step: and outputting the calculation of the layer. According to the output H of the hidden layer, connecting the weight W_jkAnd a threshold value b. The predicted output E of the network is calculated.

Where k is the number of neurons in the output layer, and m is 4.

The fourth step: and calculating the error of the network result. From E and the expected output Y, the prediction error E of the neural network is calculated.

e_k＝Y_k-E_k (3)

The fifth step: and updating the weight and the threshold of the network. Updating w according to the calculated network error_ij,w_jkAnd a_i,b_k。

w_jk＝w_jk+δH_je_k (5)

b_k＝b_k+e_k (7)

Where δ is the learning rate.

And a sixth step: and judging whether the iteration process is ended or not, and if not, returning to the second step.

For standard BP, the weights and thresholds are modified in the opposite direction of the gradient of the error function. The response rate of the network will also vary for different corrective drop algorithms. The MATLAB toolbox provides a variety of drop functions.

As shown in fig. 2, the input layer: x is the number of₁,x₂,x₃,x₄The confidences of the four models of cameras are respectively represented. Each input parameter respectively corresponds to: x is the number of₁Confidence, x, for the Unicom, model 1 camera₂Confidence, x, for the Unicom model 2 camera₃Confidence, x, for the Unicom, model 3 camera₄Is the confidence level of the contact agile 4 camera. For example, the input of {0.9,0.8.0.9,0.7} indicates that the confidence degrees of the four types of cameras when recognizing the same license plate are {0.9,0.8, 0.9,0.7}, respectively.

An output layer: y is₁,y₂,y₃,y₄The value of (d) indicates whether the camera type is selected. The type of each output should beAnd identifying the model corresponding to the more accurate camera. For example, the output is {0,1,0,0}, which indicates that camera number 2, i.e., the contact agile 2 camera, is selected. And training the weight and the threshold between the input layer and the hidden layer and between the hidden layer and the output layer by using the BP neural network to obtain the relatively accurate weight and threshold. After training, the camera recognition confidence coefficient is input every time, and the camera model with the accurate recognition result is output.

The experimental data of the neural network measured in the laboratory are shown in fig. 3.

The decimal matrix of 4 × 10 in the figure is the corresponding data of the camera confidence, and the data of 0 and 1 below is the data expected to be output. For 4 cameras to identify the same license plate, three types of cameras are required to be identified incorrectly, and the condition that one type of camera identifies correctly is very few.

For the number of hidden layers, there are three selection methods, here, equations 4-8 are chosen.

Wherein n is₁The number of the hidden layer units, n is the number of the input units, m is the number of the output units, a is [1, 10 ]]The number of

hidden layers

3,4,5,6,7 are selected as experimental parameters. And (3) constructing a neural network model by using MATLAB programming for training, wherein functions from the input layer to the hidden layer and functions from the hidden layer to the output layer adopt default functions of a BP neural network toolbox in the MATLAB, namely a tangent sigmoid transfer function tansig and a linear transfer function purelin respectively.

tansig(x)＝2/(1+exp^-2x)-1 (9)

purelin(x)＝x (10)

The number of iteration steps is set to 1000, the learning speed is set to 0.05, and the network error is set to 0.01. And finally, when the hidden layer is 5 and the Levenberg-Marquardt training algorithm is adopted for training, the iteration number and the training time are the minimum, and the training effect is the best. Input and output data and training process as shown in fig. 3 and 4, the network error reaches the requirement at step 3 of the neural network. The fitting situation is shown in fig. 5, and the fitting overall R is 0.97618. After neural network training, the correct output rate of the vehicle license plate test can reach 97%, and the main reason is that the fitting degree is good due to certain difference of the performances of the cameras, and more than 2 cameras sometimes output correct results at the same time, and the results output in the experiment are in the correct results.

For the cameras of the Haikang and Dahua types, because the recognition accuracy is very high, different calculation methods are required for some built-in development kits which do not provide confidence coefficient calculation or are related to the confidence coefficient. For a dual camera composed of different types of cameras, for example: the Haikang and the Dahua form double-camera recognition, the Haikang or the communication short forms double-camera recognition, and the Dahua and the good recognition form double-camera recognition, and because no unified comparison standard tells the system to select which camera the license plate recognized is more accurate, the system adopts a method of manually giving confidence coefficient to support any combination of 4 types of cameras, and judges which camera the recognition result should be selected according to the comparison of the manual confidence coefficient.

1) When the double camera type is formed by combining different types of communication quick cameras, the confidence coefficient value is the confidence coefficient returned by the camera recognition.

2) When the two-camera combination is made up of different types of cameras, the value of the confidence is a value read from the database according to the model of the camera and the IP of the camera. The system automatically updates the manual confidence level for that camera in the database each time the camera identifies an error.

The initial confidence is DgO, the number of vehicles with identification errors among the M vehicles identified most recently is CW, the confidence of the vehicle is Dg, and equation 51 is given.

Dg＝Dgo-CW/M (11)

In the present system test environment: m is 100, Dgo is 1.0.

The artificial confidence reflects the recent recognition rate of the camera. The following were tested in the experimental environment:

1) when 2 cameras with the identification rates which are not quite different form double-camera identification, after one camera is identified incorrectly, the next identification result of the camera is basically not used.

2) The recognition rate of the camera for common license plates (blue bottom white characters) is as high as more than 97%, the recognition errors mainly comprise license plates which are difficult to recognize, such as HK beginning, military license plates and the like, the recognition rate of the contact information is high in the common situation, but the error rate of yellow license plates is high.

Therefore, it is obviously unreasonable to select the camera with higher confidence coefficient by directly comparing the artificial confidence coefficient.

The recognition accuracy of the camera for different license plates is Correctrate, and is abbreviated as Cr. Let the recognition threshold be cr. Cr and α of Haikang and Dahua cameras in the system under the test environment are respectively 0.94 and 0.02. Setting the artificial confidence coefficient of the first camera in the double cameras as Dg₁The artificial confidence of camera two is Dg₂。

When Dg₁＞＝cr&&Dg₂When cr, the recognition degree of 2 cameras is very high, and the system selects the camera which recognizes the license plate firstly.

When Dg₁＜cr||Dg₂When the value < cr is smaller, the probability that one of 1 camera or 2 cameras in the 2 cameras have errors recently is higher, and the camera with higher artificial confidence is selected.

As shown in fig. 6, the structure stores the identification information of the camera, and is denoted as CI. Each camera IP and CI form a Key Value pair, the Key Value Key is IP, and Value is CI corresponding to the Key Value Key. In the case of dual cameras, the two different IPs correspond to the same CI.

In the invention, the time difference between the time when the camera recognizes the license plate at this time and the time when the camera recognizes the license plate at the last time is recorded as Span, and the efficiency of double-camera recognition is controlled by setting the size of the Span.

Span＝DateTime.Now()-CI.time (12)

And when the Span is 250ms, the time difference of the trigger of the double cameras is not more than 250ms, and if the time difference exceeds 250ms, the identified vehicles are not the same vehicle. And recording the currently recognized license plate, and if the Span is within 250ms, indicating that the recognition result of the camera is the result of recognizing the same license plate. By practicing the test, Span of 250ms can satisfy most situations. The effect is also better than other values.

Meanwhile, the algorithm of the heterogeneous dual-camera license plate recognition system based on confidence coefficient is started by two threads, each camera corresponds to one thread, and the recognition thread started by the camera is T₁The identification thread of the second camera is turned on is T₂The two threads share the same structural body CI respectively, and read and write are carried out on the CI when T is reached₁When reading from or writing to CI, T₂Wait until T₁After the reading and writing is finished, T₂When reading and writing the same CI, T₁Wait until T₂And finishing reading and writing.

As shown in the flow charts of fig. 7 and fig. 8, the algorithm of the confidence-based heterogeneous dual-camera license plate recognition system of the present invention comprises the following steps:

3) the algorithm ends.

In the invention, the algorithm of the fuzzy matching system for the outgoing license plate recognition comprises the following contents and processes:

through experimental tests on four types of cameras, several license plate characters which are frequently identified with errors are found, and the number of times of identifying errors is more than 10. For example, the character "2" is often recognized as the character "z".

Therefore, the present system introduces the concept of a similarity weight that reflects the probability that a camera recognizes a character s as a character t. The greater the 2-character similarity weight, the greater the probability that the camera will recognize the character s as the character t. The similarity weight of any character pair consisting of 2 characters is regulated to be 9 at most and 0 at least, the similarity weight of the character pair consisting of the same 2 characters is 9, the similarity weight of the character pair consisting of different characters is 0, and the similarity weight of the different character pairs with the similarity weight of 0 is larger at the later stage along with the increase of the number of times of error recognition of the license plate by the camera, but the maximum similarity weight is not more than 9.

Through experiments, the four cameras measure license plate character pairs which are frequently recognized wrongly and give weights to the license plate character pairs, and the larger the weight is, the higher the probability of recognition mistake between the character pairs formed by the characters s and the characters t is. A data table defined to be composed of character pairs and similarity weights corresponding to the character pairs is called a similarity dictionary. Similar dictionaries with similar weights greater than 8 are listed as shown in fig. 9.

A data table for storing similar dictionaries is arranged in a database, and the fields are as follows: character 1, character 2, similarity weight, number of errors.

The similarity dictionary increases along with the increase of the number of times of wrong license plate recognition when the vehicle leaves, and s is set to be less₁,s₂Is a similar pair of characters > is a similar pair of characters,

in order to be the weight of the similarity,

the number of errors is identified for similar characters. The algorithm flow shown in fig. 10 is:

1) obtaining a similar character and converting to a first character;

judging whether similar characters are obtained, if so, turning to 2, otherwise, turning to the next step;

judging whether the similar character pair is in the dictionary, if so, turning to the third step, otherwise, turning to the sixth step;

judging whether the similarity weight is equal to 9, if so, turning to 2), and otherwise, turning to the fourth step;

fourthly, reading the database and judging the times of secondary errors

In the case of a FALSE,

and writing into the database. Turning to 2), if TRUE,

fifthly, turning;

fifthly, adding 1 to the similarity weight of the similar character, namely

And adds 1 to the number of errors. And updating the database and updating the data dictionary. Turning 2);

sixthly, the character pair is less than s₁,s₂> (similar) weight

Number of errors

And inserting into a database. Turning 2);

2) the algorithm ends.

Recording the similarity as D, P as the license plate recognized by the camera, T as the license plate to be matched, and the fuzzy matching threshold value as W and x_iIs the ith character of P, y_iThe ith character of T, d is the similarity weight of the character pair and d ═ f (x)_iy_j) And L is the length of the license plate, and the similarity formula is shown as 1.

If D is larger than or equal to the given fuzzy matching threshold value W, P, T is regarded as the same license plate, otherwise, the license plate is regarded as a different license plate. The fuzzy matching threshold value W is selected according to environment and the quality of camera identification, and can be taken as W₁,W₂,W₃. M is the maximum weight 9, s is the number of allowed fuzzy characters, and fa is the adjustment factor. The threshold calculation formulas are shown in fig. 14,15 and 16.

W₁＝L*M (14)

W₂＝(L-s)*M+s*fa fa∈(0,M]And fa ∈ N (15)

W₃＝(L-s)*M (16)

Wherein W takes the value W₁Accurate matching is said to not allow the license plate to recognize the wrong characters. Value W₂Referring to similarity fuzzy matching, we allow P, T to have s character differences, but the character differences should not deviate from the character dictionary, and the larger fa, the greater the similarity weight between different characters is required. Value W₃Referring to normal fuzzy matching, P, T are allowed to have s number of different characters, and the similarity weight of different characters may be 0.

The similarity is directly calculated by using a similarity formula to compare fuzzy matching threshold values to judge whether the P is matched or not, and the algorithm for judging whether the T is matched is 'general fuzzy matching'. The condition that the type of the license plate recognition error is P and the recognition error on the corresponding position of T can be solved by the fuzzy matching, and the condition that the license plate recognition is staggered can not be solved.

In the invention, the camera provides fuzzy matching based on the longest public subsequence based on the fuzzy matching principle for the condition that the license plate recognition has recognition dislocation, and does not directly use a similarity formula to calculate the similarity to compare fuzzy matching threshold values. For P, T is matched according to the method of the longest common subsequence, and whether the similarity weight of the character pair is larger than or equal to a specified threshold dt is judged. And D (i, j) is the maximum similarity between the first i character strings of P and the first j character strings of T, and the recursive formula relation of P based on the longest common subsequence is shown as 17.

When D (L, L) > -, W indicates that the license plates are the same, and the matching is successful; when D (L, L) < W indicates that different license plates are identified, the matching fails. The recursion formula uses a dynamic programming algorithm principle to reduce the time complexity of one-time matching of the staggered license plate from O (2^ L ^ 2^ L) to O (L ^ 2). If n vehicles are arranged in the parking lot, the complexity of license plate matching is O (L ^2 x n); the fuzzy matching based on the longest common subsequence is called fuzzy matching based on a similar dictionary, and is abbreviated as fuzzy matching. The fuzzy matching is accurate, common and similar fuzzy matching according to fuzzy matching threshold selection, each fuzzy matching firstly calls general fuzzy matching to match the license plate, and when the general fuzzy matching is not used for matching the license plate in the field, the system automatically uses the fuzzy matching based on the longest public word sequence to match the license plate in the field. The upper time complexity limit for fuzzy matching is therefore O (n L + L2 n). Since recognition misplacement is rare in a test environment, the temporal complexity of fuzzy matching is typically O (L × n).

As shown in fig. 11, the fuzzy matching algorithm flow of the departure license plate recognition is as follows:

1) acquiring a license plate P identified by a camera;

judging that i is equal to N, and if i is FALSE, turning to the fifth step,

if TRUE, go to sixteenth;

Otherwise, turning to the seventh step;

Otherwise, turning to manual processing and turning to 3);

license plate T [ x ] with maximum similarity]Go to 3);

3) the algorithm ends.

In the fuzzy matching algorithm for the outgoing license plate recognition, the similarity is D, P is the license plate recognized by the camera, T is the license plate to be matched, the fuzzy matching threshold is W, x_iIs the ith character of P, y_iThe ith character of T, d is the similarity weight of the character pair and d ═ f (x)_iy_j) And L is the length of the license plate.

The fuzzy matching design aims to reduce manual intervention and improve the intelligence degree of the parking lot. Herein, a concept of the degree of intelligence available for computation is defined. When the license plate recognition rate is 100%, the number of registered vehicles that should be automatically discharged is CAuto, and is abbreviated as CA. In the practical situation, the number of the vehicles which should be automatically discharged from the vehicle registration system and need manual intervention due to wrong license plate recognition is recorded as CManual, which is abbreviated as CM; the number of vehicles which can not be automatically taken out from the registered vehicles due to other problems is called "common" and abbreviated as COM, the vehicles which are taken out from the vehicle in the 2 cases are called "forced manual vehicles", and the Intelligence degree is denoted as Intelligence level and abbreviated as ITL. There is the formula:

ITL＝(CA-CM-COM)/CA (18)

when the ITL is 1, it means that all registered and non-defaulting vehicles can automatically enter and exit without manual intervention. Where the size of the CM is mainly determined by hardware conditions, the CM may be reduced using dual cameras in case of camera type determination. The COM is influenced by many factors, such as sudden network failure of a parking lot, and some registered vehicles cannot be automatically delivered. The shortage of the balance of the recharging vehicle results in that the registered vehicle can not automatically go out of the field. The offline processing and charging processing discussed in detail below may reduce the COM value appropriately, increasing the intelligence of the parking lot.

The fuzzy matching algorithm can reduce the number of forced manual vehicles, so that the license plate can be automatically brought out under the condition of camera recognition error, and the number of the vehicles subjected to fuzzy matching processing is recorded as CSimilar, which is abbreviated as CS. Comprises the following steps:

ITL＝(CA-COM-(CM-CS))/CA (19)

the fuzzy matching algorithm is used for realizing automatic departure of the vehicles, the fuzzy matching is required to return only one unique license plate under the condition of wrong identification, but with the increase of the vehicles in the parking lot, the ordinary fuzzy matching is used for returning the unique license plate possibly, at the moment, the similar fuzzy matching is used for improving the precision of the fuzzy matching, and therefore the size of the CS is related to the number of the vehicles in the parking lot and the mode of the fuzzy matching. The number of vehicles registered in the field is CIP, and the fuzzy matching mode is W_i. Therefore, there are:

CS＝F(CIP,W_i)i＝{1,2,3} (20)

ITL＝f(CA,CM,COM,CIP,W_i)i＝{1,2,3} (21)

in a laboratory environment, a Dahua camera and a Haikang camera are used for testing, 5000 vehicles of license plates are entered and exited for cycle testing (the license plates have repeated conditions and contain special license plates which are not easily identified, such as HK head, military license plates and the like), and the test data of common fuzzy matching is used under the condition of single-camera identification, and is shown in figure 12.

When there are more than a certain number of vehicles present, using ordinary fuzzy matching often returns results of more than 1 vehicle after recognition error, resulting in an inability to automatically go out of the field. And (4) testing the same 5000 license plates, and changing to similar fuzzy matching under the condition that the identification results of multiple times of common fuzzy matching return more than 1 license plate. The test data are shown in fig. 13.

With the fuzzy matching algorithm, the intelligence level can be improved by at least 3 percentage points in a laboratory environment, i.e., COM is 0.

As shown in fig. 14, Dahua 105 lane, 2 cameras with different types are arranged, one is a KD-XJ02 camera, and the other is a KD-XJ03 camera. When the auxiliary camera parameter type is KD-XJ05, ETC is represented; as fig. 15 corresponds to the camera parameter configuration of 5.1, 2 different types of cameras are monitored independently in the same inbound lane. Only one camera is used for monitoring the outgoing lane independently. An entrance lane in the monitoring system uses an algorithm of a heterogeneous dual-camera system based on confidence to carry out dual-camera monitoring, and when the identification of the outgoing license plate is wrong, the system automatically calls the algorithm of the fuzzy matching system to match the existing license plate.

It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential attributes thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.

Claims

1. The intelligent control method for the access of the parking lot is characterized by comprising the following steps:

3) a heterogeneous dual-camera license plate recognition system based on confidence coefficient in the monitoring system recognizes the data acquired in the step 2), wherein the confidence coefficient comprises camera confidence coefficient, the camera confidence coefficient is the judgment degree of the accuracy of the license plate recognized by the camera, and the heterogeneous dual-camera license plate recognition system based on the confidence coefficient comprises a recognition result of which camera is judged according to the camera confidence coefficient;

9) and sending the charging information obtained by the monitoring system to a charging system, charging by the charging system, feeding the charged information back to the monitoring system, and finishing monitoring.

2. The intelligent parking lot access control method according to claim 1, wherein: the algorithm of the heterogeneous dual-camera license plate recognition system based on the confidence coefficient comprises the following steps:

1) acquiring camera parameters of a corresponding lane, initializing CI, starting camera identification threads T1 and T2, and waiting for triggering;

obtaining a license plate P and a confidence coefficient Dg, updating a structure CI, wherein the CI is CI.last _ plate is P, the CI.last _ plate is CI.plate, the CI.dg is Dg, modifying license plate information in a database, and turning to 3);

3) the algorithm ends.

3. The intelligent parking lot access control method according to claim 2, wherein: the Span is the time difference between the time when the camera recognizes the license plate at this time and the time when the camera recognizes the license plate at the last time, the algorithm of the heterogeneous dual-camera license plate recognition system based on the confidence coefficient is started by two threads, each camera corresponds to one thread, the recognition thread started by the first camera is T1, the recognition thread started by the second camera is T2, the two threads share the same structural body CI respectively, the CI is read and written, when the T1 reads and writes the CI, the T2 waits until the T1 finishes reading and writing, and when the T2 reads and writes the same CI, the T1 waits until the T2 finishes reading and writing.

4. The intelligent parking lot access control method according to any one of claims 1 to 3, wherein: the fuzzy matching system for the departure license plate recognition comprises the following algorithm flows:

1) acquiring a license plate P identified by a camera;

fifthly, calculating the similarity D of the two license plates, if D is greater than W, recording the license plate T [ i ], and turning to the fourth, otherwise, directly turning to the fourth;

sixthly, judging whether similar license plate records exist or not, if the license plate records are TRUE, taking out the only license plate T [ x ], calling a dictionary-based self-learning algorithm to calculate similar character pairs, and if not, turning to the step (c);

Otherwise, turning to manual processing and turning to 3);

number plate T [ x ] with maximum return similarity]Go to 3);

3) the algorithm ends.

5. The intelligent parking lot access control method according to claim 4, wherein: in the algorithm of the fuzzy matching system for the outgoing license plate recognition, the similarity is D, P is the license plate recognized by a camera, T is the license plate to be matched, the fuzzy matching threshold is W, x i is the ith character of P, y i is the ith character of T, D is the similarity weight of a character pair, D is f (x i y j), and L is the length of the license plate.

6. A parking lot access intelligent control system to which the parking lot access intelligent control method according to any one of claims 1 to 5 is applied, characterized in that: the system comprises a first camera, a second camera, a third camera, a monitoring system and a charging system, wherein the first camera and the second camera are arranged at an entrance of a parking lot, the third camera is arranged at an exit of the parking lot, the first camera, the second camera, the third camera and the charging system are all connected with the monitoring system, and the monitoring system comprises a heterogeneous dual-camera license plate recognition system based on confidence and a fuzzy matching system for license plate recognition of a departure place.

7. The intelligent parking lot access control system according to claim 6, wherein: the intelligent control system further comprises a charging system connected with the monitoring system, and the charging system is an ETC system.