CN109034211A - A kind of parking space state detection method based on machine learning - Google Patents
A kind of parking space state detection method based on machine learning Download PDFInfo
- Publication number
- CN109034211A CN109034211A CN201810721730.9A CN201810721730A CN109034211A CN 109034211 A CN109034211 A CN 109034211A CN 201810721730 A CN201810721730 A CN 201810721730A CN 109034211 A CN109034211 A CN 109034211A
- Authority
- CN
- China
- Prior art keywords
- parking stall
- picture
- machine learning
- model
- parking
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N7/00—Television systems
- H04N7/18—Closed-circuit television [CCTV] systems, i.e. systems in which the video signal is not broadcast
Abstract
The present invention discloses a kind of parking space state detection method based on machine learning, which comprises step A, collecting cart bitmap piece, the parking stall picture include empty wagons parking stall picture and non-empty wagon parking stall picture;Step B handles the parking stall picture of collection;Step C, to treated, picture is classified;Step D, establishes machine learning model;Step E is trained the machine learning model of foundation;Step F tests the machine learning model after training;Step G, to by test after machine learning model carry out using.Parking space state detection method provided by the invention based on machine learning has the advantage that parking stall need to only install common camera, does not need high definition camera, can reduce equipment purchasing and maintenance cost.One video camera can detect multiple parking stalls, install a video camera without each parking stall.To vehicle parking position without strict demand, vehicle need to only stop parking stall.
Description
Technical field
The present invention relates to vehicle parking automation control area more particularly to a kind of parking space state inspections based on machine learning
Survey method.
Background technique
Parking lot vehicle guide system be guide vehicle fast running arrive available free parking stall system, at present create in greatly
Generally there is installation in type parking garage.Its basis is planning and identification to idle parking stall.Current general parking lot vehicle
Whether guidance system is idle using following several method identification parking stall: 1, earth induction device.It is installed in parking stall.Using electricity
Whether magnetic induction principle, detection parking stall are parked vehicle.2, ultrasonic detection device.It is installed on above parking stall.Using ultrasonic reflection
It is detected at a distance from ground, whether vehicle is parked by comparing distance detection parking stall.3, vehicle/license plate recognition device.Installation
In parking stall oblique upper.Vehicle and license plate in intelligent video camera head identification shooting photo, to judge whether parking stall is parked vehicle.
But above-mentioned parking lot vehicle guide system with the following drawback that:
1, earth induction device.It need to be installed in parking stall, have destruction to ground, vehicle need to be standardized and be parked, otherwise may sense
It should be less than induction speed is relatively slow, feedback transmission speed is slower, and a parking stall needs a device, therefore original equipment cost and peace
It is higher to fill maintenance expense.
2, ultrasonic detection device.Vehicle need to be standardized and be parked, and otherwise may can't detect, more sensitive to height of car,
Therefore be only applicable to certain a kind of vehicle, a parking stall needs a device, therefore original equipment cost and installation maintenance take it is higher.
3, vehicle/license plate recognition device.It need to be installed on vehicle diagonally forward, to the height of video camera, video camera and parking stall
Distance and angle require, and vehicle need to be standardized and be parked, otherwise may identification less than, the dynamic adaptable of Car license recognition is not high,
Discrimination such as Novel license plate, foreign license plate is lower.Video camera needs high definition and can identify license plate, therefore original equipment cost and installation
Maintenance expense is high.
In consideration of it, need to design a kind of parking space state detection method based on machine learning, it can be simple using the method
Dead ship condition that is convenient, accurately judging parking stall.
Summary of the invention
The present invention aiming at the problems existing in the prior art, provides a kind of parking space state detection side based on machine learning
Method, which comprises
Step A, collecting cart bitmap piece, the parking stall picture include empty wagons parking stall picture and non-empty wagon parking stall picture;
Step B handles the parking stall picture of collection;
Step C, to treated, picture is classified;
Step D, establishes machine learning model;
Step E is trained the machine learning model of foundation;
Step F tests the machine learning model after training;
Step G, to by test after machine learning model carry out using.
Further, the method also includes:
Step H is modified the machine learning model used.
Further, in the step A collect parking stall picture according to camera numbers, whether there is vehicle to classify.
Further, the step B is specifically included: being deleted to the image of non-parking stall part in the parking stall picture being collected into
It removes, leaves the image of parking stall part, to including multiple parking stall images in the parking stall picture being collected into, bicycle position is carried out to picture
Segmentation, so that each width picture is only reflected a parking stall, and to picture after treatment by parking stall numbering, whether have vehicle progress
Classification.
Further, the step C is specifically included: to each parking stall, by the 90% of its empty parking space picture number as instruction
Practice data set Q, is used as validation data set R for the 10% of empty parking space picture number, there will be vehicle picture as test data set Z.
Further, the step D is specifically included: establishing machine learning mould using convolutional neural networks CNN algorithm
Type, the convolutional neural networks CNN algorithm include data processing relevant parameter, training process and training relevant parameter, network phase
Close parameter and classifier decision threshold parameter M.
Further, the step E is specifically included: using parking stall data set Q as input, training empty parking space picture feature is known
Other model.For model loss function to export 1 as minimum, closer 1 is better.But realistic model output valve is considered as when being greater than M can
Receive.The verifying of model uses data set R.With data set R input model, output is excellent close to 1.But realistic model output valve
It is considered as when greater than M and is verified, when model output is acceptable, model training terminates.
Further, the step F is specifically included: the model exported with data set Z test model training step.With output
It is excellent close to 0.But realistic model output valve is considered as test when being less than (1-M) and passes through.
Further, step G is specifically included: when using model, the parking stall picture after segmentation is inputted to the model of the parking stall,
Current parking space state is judged according to its output valve V, works as V >=M, then judges that parking stall for idle state, if V≤1-M, judges vehicle
Position is that occupied state is then modified model as 1-M < V < M.
Further, the step H is specifically included: the input using the picture subset acquired recently as training pattern,
Time picture remote directly abandons, and initial parameter adjustment was carried out on the basis of former model, and multiplexing parameters periodically carry out mould
Type amendment.
Compared with prior art, the parking space state detection method provided by the invention based on machine learning has following excellent
Point:
1, parking stall need to only install common camera.Do not need high definition, do not need to identify license plate, can reduce equipment purchasing and
Maintenance cost.
2, installation site no requirement (NR).It is mounted on parking stall top or diagonally forward.
3, a video camera can detect multiple parking stalls, install a video camera without each parking stall.
4, to vehicle parking position without strict demand, vehicle need to only stop parking stall.
5, to vehicle no requirement (NR).Suitable for a variety of models, all kinds of license plates.
6, the accuracy of judgement degree of parking space state can increase continuous improvement with image data.
Detailed description of the invention
Fig. 1 is the flow chart of the parking space state detection method of the invention based on machine learning.
Specific embodiment
With reference to the accompanying drawings and examples, specific embodiments of the present invention will be described in further detail.Implement below
Example is not intended to limit the scope of the invention for illustrating the present invention.
As shown in Figure 1, for the present invention is based on a kind of embodiment of the parking space state detection method of machine learning, this method packets
It includes:
Step A, collecting cart bitmap piece, the parking stall picture include empty wagons parking stall picture and non-empty wagon parking stall picture;
Step B handles the parking stall picture of collection;
Step C, to treated, picture is classified;
Step D, establishes machine learning model;
Step E is trained the machine learning model of foundation;
Step F tests the machine learning model after training;
Step G, to by test after machine learning model carry out using.
Step H is modified the machine learning model used.
Wherein, in the step A, after the installation of parking stall camera, timing acquiring parking stall picture to backstage.The picture of acquisition will be by
Whether camera number has vehicle to classify.
Wherein, in the step B, parking stall photo also contains non-parking stall part generally in addition to the image containing parking stall part
Image, such as aisle, column, parking stall surrounding ground.The picture to parking stall part is needed to intercept, to accurately reflect parking stall figure
Picture.One multiple parking stall of camera, also the segmentation of bicycle position is carried out to picture, each width figure be made only to reflect a parking stall, and
To picture after treatment by parking stall numbering, whether there is vehicle to classify.
Wherein, in the step C, to each parking stall, it is used as training dataset Q by the 90% of its empty parking space picture number,
It is used as validation data set R by the 10% of empty parking space picture number, there will be vehicle picture as test data set Z.
Wherein, in step D, the model that each parking stall needs to establish is different, but its modeling method is consistent.
Backstage parking space management system establishes machine learning model, and convolutional neural networks CNN generally can be used.Tentatively draft with
Lower parameter:
Data processing (or pretreatment) relevant parameter.Such as data enhancing, extensive processing, BN processing.
Training process parameter relevant to training.Such as gradient descent algorithm, the number of iterations, learning rate, attenuation function, weight
Initialization, regularization correlation technique etc..
Network-related parameters.As how the network number of plies, every layer of neuron number, number of filters and classifier select.
Classifier judgment threshold M.It indicates the result output 1 of parking stall picture input model, calculating " to be determined as empty wagons completely
Position ", output 0 indicates " determining not to be empty parking space completely ", if output valve P between 0 and 1, means that the parking stall for sky
The probability of parking stall is P.If P >=M, is classified as empty parking space, it is otherwise classified as non-empty parking stall.Noting: general M value answers >=
0.8.
Wherein, in step E, model training is carried out respectively to each parking stall.Using parking stall data set Q as input, training is empty
Parking stall picture feature identification model.For model loss function to export 1 as minimum, closer 1 is better.But realistic model output valve is big
It is considered as when M acceptable.The verifying of model uses data set R.With data set B input model, output is excellent close to 1.But
Realistic model output valve is considered as when being greater than M to be verified.When model output is acceptable, training comes to an end.Model enters
Test phase.
Wherein, in step F, with the model of data set Z test model training step output.With output close to 0 to be excellent.But
Realistic model output valve is considered as test when being less than (1-M) and passes through.
Wherein, in step G, after testing the model passed through, empty parking space picture and non-empty wagon bitmap piece by model calculating
The difference of output valve is more than 2* (M-0.5), it is sufficient to distinguish parking space state.Such as M=0.8, then model distinguishes two kinds
The section of parking space state picture is respectively: empty parking space photographic model output valve is located at section: (0.8,1), non-empty wagon bitmap piece mould
Type output valve is located at section: (0,0.2), therefore, when using model, the parking stall picture after segmentation is inputted to the model of the parking stall,
It is easy for judging current parking space state according to its output valve V.
If V >=0.8, parking stall is idle state
If V≤0.2, parking stall is occupied state
If V is between 0.2 and 0.8, parking stall is judged as occupied state.But model need to be modified.
Wherein, in step H, model reflection be empty wagons bit image under the conditions of a certain period feature, when feature becomes
When change, model is also required to be modified.For example the daily illumination of certain parking stall has become, has been mounted with a car lug, re-starts
Scribing line etc., empty parking space characteristics of image model just needs to correct.The method of Modifying model is as follows:
Input using the picture subset acquired recently as training pattern.Time picture remote directly abandons, and can subtract
It interferes less and shortens the modeling time.
Initial parameter adjustment was carried out on the basis of former model.Multiplexing parameters can accelerate modeling speed.
Modifying model can be carried out, actively periodically with the applicability of assurance model.
The present invention provides a kind of parking space state detection method based on machine learning, and this method uses machine learning model,
Whether automatization judgement, which stops, is carried out to parking space state, simple and convenient, accuracy rate is high, be provided simultaneously with following advantages: parking stall only needs
Common camera is installed.It does not need high definition, need to identify license plate, equipment purchasing and maintenance cost can be reduced.Installation site
No requirement (NR).It is mounted on parking stall top or diagonally forward.One video camera can detect multiple parking stalls, without each parking stall installation one
A video camera.To vehicle parking position without strict demand, vehicle need to only stop parking stall.To vehicle no requirement (NR).Suitable for each
Kind vehicle, all kinds of license plates.The accuracy of judgement degree of parking space state can increase continuous improvement with image data.
The foregoing is merely presently preferred embodiments of the present invention, is not intended to restrict the invention, it is all in spirit of the invention and
In principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (10)
1. a kind of parking space state detection method based on machine learning, which is characterized in that the described method includes:
Step A, collecting cart bitmap piece, the parking stall picture include empty wagons parking stall picture and non-empty wagon parking stall picture;
Step B handles the parking stall picture of collection;
Step C, to treated, picture is classified;
Step D, establishes machine learning model;
Step E is trained the machine learning model of foundation;
Step F tests the machine learning model after training;
Step G, to by test after machine learning model carry out using.
2. the parking space state detection method according to claim 1 based on machine learning, which is characterized in that the method is also
Include:
Step H is modified the machine learning model used.
3. the parking space state detection method according to claim 1 based on machine learning, which is characterized in that the step A
It is middle collect parking stall picture according to camera numbers, whether there is vehicle to classify.
4. the parking space state detection method according to claim 1 based on machine learning, which is characterized in that the step B
It specifically includes: the image of non-parking stall part in the parking stall picture being collected into being deleted, the image of parking stall part is left, to receipts
Comprising multiple parking stall images in the parking stall picture collected, the segmentation of bicycle position is carried out to picture, each width picture is made only to reflect one
A parking stall, and to picture after treatment by parking stall numbering, whether there is vehicle to classify.
5. the parking space state detection method according to claim 1 based on machine learning, which is characterized in that the step C
It specifically includes: to each parking stall, training dataset Q is used as by the 90% of its empty parking space picture number, by empty parking space picture number
10% be used as validation data set R, will have vehicle picture as test data set Z.
6. the parking space state detection method according to claim 5 based on machine learning, which is characterized in that the step D
It specifically includes: machine learning model, the convolutional neural networks CNN algorithm is established using convolutional neural networks CNN algorithm
Join including data processing relevant parameter, training process and training relevant parameter, network-related parameters and classifier decision threshold
Number M.
7. the parking space state detection method according to claim 6 based on machine learning, which is characterized in that the step E
It specifically includes: using parking stall data set Q as input, training empty parking space picture feature identification model.Model loss function is to export 1
It is better closer to 1 for minimum.But realistic model output valve is considered as acceptable when being greater than M.The verifying of model uses data set R.
With data set R input model, output is excellent close to 1.But realistic model output valve is considered as when being greater than M to be verified, and mould is worked as
When type output is acceptable, model training terminates.
8. the parking space state detection method according to claim 7 based on machine learning, which is characterized in that the step F
It specifically includes: the model exported with data set Z test model training step.With output close to 0 to be excellent.But realistic model exports
Value is considered as test when being less than (1-M) and passes through.
9. the parking space state detection method according to claim 8 based on machine learning, which is characterized in that step G is specific
When including: using model, the parking stall picture after segmentation is inputted to the model of the parking stall, current parking stall is judged according to its output valve V
State works as V >=M, then judges parking stall for idle state, if V≤1-M, judges parking stall for occupied state, as 1-M < V < M,
Then model is modified.
10. the parking space state detection method according to claim 2 based on machine learning, which is characterized in that the step H
Specifically include: the input using the picture subset that acquires recently as training pattern, time picture remote directly abandon, with
Initial parameter adjustment is carried out on the basis of preceding model, multiplexing parameters periodically carry out Modifying model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810721730.9A CN109034211A (en) | 2018-07-04 | 2018-07-04 | A kind of parking space state detection method based on machine learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810721730.9A CN109034211A (en) | 2018-07-04 | 2018-07-04 | A kind of parking space state detection method based on machine learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109034211A true CN109034211A (en) | 2018-12-18 |
Family
ID=65522236
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810721730.9A Pending CN109034211A (en) | 2018-07-04 | 2018-07-04 | A kind of parking space state detection method based on machine learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109034211A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109635768A (en) * | 2018-12-20 | 2019-04-16 | 深圳市捷顺科技实业股份有限公司 | Parking stall condition detection method, system and relevant device in a kind of picture frame |
CN109817013A (en) * | 2018-12-19 | 2019-05-28 | 新大陆数字技术股份有限公司 | Parking stall state identification method and device based on video flowing |
CN110852313A (en) * | 2020-01-15 | 2020-02-28 | 魔视智能科技(上海)有限公司 | Parking space detection method |
CN114694124A (en) * | 2022-05-31 | 2022-07-01 | 成都国星宇航科技股份有限公司 | Parking space state detection method and device and storage medium |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102663357A (en) * | 2012-03-28 | 2012-09-12 | 北京工业大学 | Color characteristic-based detection algorithm for stall at parking lot |
CN104112370A (en) * | 2014-07-30 | 2014-10-22 | 哈尔滨工业大学深圳研究生院 | Monitoring image based intelligent parking lot parking place identification method and system |
CN105225523A (en) * | 2015-10-15 | 2016-01-06 | 浙江宇视科技有限公司 | A kind of parking space state detection method and device |
US20180025640A1 (en) * | 2016-07-19 | 2018-01-25 | Ford Global Technologies, Llc | Using Virtual Data To Test And Train Parking Space Detection Systems |
CN107886080A (en) * | 2017-11-23 | 2018-04-06 | 同济大学 | One kind is parked position detecting method |
CN108108689A (en) * | 2017-12-18 | 2018-06-01 | 北京工业大学 | A kind of intelligent parking method for detecting parking stalls |
-
2018
- 2018-07-04 CN CN201810721730.9A patent/CN109034211A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102663357A (en) * | 2012-03-28 | 2012-09-12 | 北京工业大学 | Color characteristic-based detection algorithm for stall at parking lot |
CN104112370A (en) * | 2014-07-30 | 2014-10-22 | 哈尔滨工业大学深圳研究生院 | Monitoring image based intelligent parking lot parking place identification method and system |
CN105225523A (en) * | 2015-10-15 | 2016-01-06 | 浙江宇视科技有限公司 | A kind of parking space state detection method and device |
US20180025640A1 (en) * | 2016-07-19 | 2018-01-25 | Ford Global Technologies, Llc | Using Virtual Data To Test And Train Parking Space Detection Systems |
CN107886080A (en) * | 2017-11-23 | 2018-04-06 | 同济大学 | One kind is parked position detecting method |
CN108108689A (en) * | 2017-12-18 | 2018-06-01 | 北京工业大学 | A kind of intelligent parking method for detecting parking stalls |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109817013A (en) * | 2018-12-19 | 2019-05-28 | 新大陆数字技术股份有限公司 | Parking stall state identification method and device based on video flowing |
CN109635768A (en) * | 2018-12-20 | 2019-04-16 | 深圳市捷顺科技实业股份有限公司 | Parking stall condition detection method, system and relevant device in a kind of picture frame |
CN109635768B (en) * | 2018-12-20 | 2023-02-28 | 深圳市捷顺科技实业股份有限公司 | Method and system for detecting parking space state in image frame and related equipment |
CN110852313A (en) * | 2020-01-15 | 2020-02-28 | 魔视智能科技(上海)有限公司 | Parking space detection method |
CN110852313B (en) * | 2020-01-15 | 2020-04-17 | 魔视智能科技(上海)有限公司 | Parking space detection method |
CN114694124A (en) * | 2022-05-31 | 2022-07-01 | 成都国星宇航科技股份有限公司 | Parking space state detection method and device and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109034211A (en) | A kind of parking space state detection method based on machine learning | |
CN112380952B (en) | Power equipment infrared image real-time detection and identification method based on artificial intelligence | |
CN106652468B (en) | The detection and from vehicle violation early warning alarm set and method in violation of rules and regulations of road vehicle front truck | |
CN103235933B (en) | A kind of vehicle abnormality behavioral value method based on HMM | |
CN102076531B (en) | Vehicle clear path detection | |
CN109816024A (en) | A kind of real-time automobile logo detection method based on multi-scale feature fusion and DCNN | |
CN110288823B (en) | Traffic violation misjudgment identification method based on naive Bayesian network | |
CN111079640B (en) | Vehicle type identification method and system based on automatic amplification sample | |
KR101973933B1 (en) | Method and Apparatus for Detecting Boarding Number | |
CN109191830A (en) | A kind of congestion in road detection method based on video image processing | |
CN109344886B (en) | Occlusion number plate distinguishing method based on convolutional neural network | |
CN113034378B (en) | Method for distinguishing electric automobile from fuel automobile | |
CN110209999A (en) | A kind of mobile unit failure trend prediction method | |
CN117173913B (en) | Traffic control method and system based on traffic flow analysis at different time periods | |
CN109447090A (en) | A kind of shield door obstacle detection method and detection system | |
CN114049612A (en) | Highway vehicle charge auditing system based on graph searching technology and dual-obtaining and checking method for driving path | |
CN112381014A (en) | Illegal parking vehicle detection and management method and system based on urban road | |
CN111862065A (en) | Power transmission line diagnosis method and system based on multitask deep convolutional neural network | |
CN109657580A (en) | A kind of urban track traffic gate passing control method | |
CN113378646A (en) | Freight train information identification system and identification method | |
CN116631186A (en) | Expressway traffic accident risk assessment method and system based on dangerous driving event data | |
KR102561656B1 (en) | Control system of traffic flow with sensing of vehicle based on deep learning | |
CN109344829B (en) | Train number identification method and device for high-speed railway train | |
CN106446929B (en) | Type of vehicle detection method based on edge gradient potential energy | |
CN116597411A (en) | Method and system for identifying traffic sign by unmanned vehicle in extreme weather |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181218 |