CN107067813A - A kind of parking stall bootstrap technique and system based on image procossing and pattern-recognition - Google Patents
A kind of parking stall bootstrap technique and system based on image procossing and pattern-recognition Download PDFInfo
- Publication number
- CN107067813A CN107067813A CN201710447805.4A CN201710447805A CN107067813A CN 107067813 A CN107067813 A CN 107067813A CN 201710447805 A CN201710447805 A CN 201710447805A CN 107067813 A CN107067813 A CN 107067813A
- Authority
- CN
- China
- Prior art keywords
- mark rod
- parking stall
- mark
- vehicle
- coordinate
- 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
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/14—Traffic control systems for road vehicles indicating individual free spaces in parking areas
- G08G1/141—Traffic control systems for road vehicles indicating individual free spaces in parking areas with means giving the indication of available parking spaces
- G08G1/142—Traffic control systems for road vehicles indicating individual free spaces in parking areas with means giving the indication of available parking spaces external to the vehicles
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
- G06V20/586—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of parking space
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of parking stall bootstrap technique and system based on image procossing and pattern-recognition, including：Image to be detected of camera collection is received, described image to be detected includes：Mark rod and multiple parking stalls, the mark rod are arranged on four summits of the multiple parking stall correspondence rectangular area or cornerwise two points；Mark rod in the identification parking field picture, and position of the mark rod in the parking stall image is determined according to the coordinate of the mark rod；The vehicle in the range of the mark rod is recognized, and determines the number of the vehicle；Compare the vehicle number and the parking stall number in the scope, if the number of the vehicle is equal to the parking stall number, it is determined that without empty parking space, if the number of the vehicle is less than the parking stall number, the position of empty parking space is determined according to the coordinate of the vehicle.The present invention uses existing monitoring camera combination mark rod, realizes parking stall guiding based on image procossing and pattern-recognition, reduces the cost of system, beneficial to later maintenance, debugging and upgrading.
Description
Technical field
Draw the present invention relates to parking stall bootstrap technique field, more particularly to a kind of parking stall based on image procossing and pattern-recognition
Guiding method and system.
Background technology
With the increasing number of automobile, the problem of generating parking difficulty therewith.Many vehicles, which enter to encounter behind garage, to be sought
Look for the problem of parking stall.
The conventional use ultrasonic sensor of current truck space guiding system, the sensor such as magnetic detector and according to its from
The operation principle of body, is monitored in real time to the parking space state on each parking stall in parking lot, and statistics parking position is used
Condition information.In recent years, the method for carrying out single-point detection to parking stall using camera is also occurred in that, single-point detection mode has inspection
The advantages of data are accurate, real-time is high is surveyed, domestic most of parking lots use this data acquisition modes.
But aforesaid way causes cost higher, and construction period is longer.It is larger in the presence of difficulty is transformed existing parking lot,
The shortcomings of being unfavorable for later maintenance, debugging simultaneously and upgrade.
The content of the invention
The present invention provides a kind of parking stall bootstrap technique and system based on image procossing and pattern-recognition, to overcome above-mentioned skill
Art problem.
A kind of parking stall bootstrap technique based on image procossing and pattern-recognition of the present invention, including：
Image to be detected of camera collection is received, described image to be detected includes：Mark rod and multiple parking stalls, the mark
Know bar and be arranged on four summits of the multiple parking stall correspondence rectangular area or cornerwise two points；
Mark rod in the identification parking field picture, and determine the mark rod in institute according to the coordinate of the mark rod
State the position in the image of parking stall；
The vehicle in the range of the mark rod is recognized, and determines the number of the vehicle；
Compare the vehicle number and the parking stall number in the scope, if the number of the vehicle is equal to the parking stall number,
No empty parking space is then determined, if the number of the vehicle is less than the parking stall number, empty wagons is determined according to the coordinate of the vehicle
The position of position.
Further, the mark rod in the identification parking field picture, and being determined according to the coordinate of the mark rod
After position of the mark rod in the parking stall image, in addition to：
Judge whether the number of mark rod is less than 2, if so, then alarm signal is sent to control centre, if it is not, then judging institute
Whether the number for stating mark rod is equal to 2；
If the mark rod number is 2, judge whether two mark rods are in diagonally according to the coordinate of the mark rod
On line, if it is not, alarm signal then is sent to control centre, if, it is determined that described two mark rods are effective；
If the mark rod number is not 2, the mark rod on diagonal is chosen according to the coordinate of the mark rod.
Further, the mark rod in the parking field picture is recognized, and according to being determined the coordinate of the mark rod
Position of the mark rod in the parking stall image, including：
Mark rod data set is obtained from picture of the database comprising mark rod；
Convolutional neural networks model is built, and the model is trained according to the data set；
The mark rod is detected using the model after training and described image to be detected, the mark rod center is stored
The coordinate of point.
Further, the vehicle in the range of the identification mark rod, including：
The central point for connecting the mark rod determines the scope of parking stall measure；
Vehicle data collection is obtained from picture of the database comprising vehicle, convolutional neural networks model is built, and according to institute
State data set and train the model；
The vehicle is detected using image to be detected in the range of the model and parking stall measure after training.
Further, it is described, the position of empty parking space is determined according to the coordinate of the vehicle, including：
The corresponding rectangle of the mark rod is determined according to the coordinate of two mark rods on the diagonal, according to the square
Parking stall number in shape divides equally the rectangle for several parking stall rectangles；
The position of empty parking space is determined according to the coordinate of the vehicle and the parking stall rectangle.
Further, after the coordinate according to the vehicle and parking stall rectangle determine the position of empty parking space, go back
Including：
The position of the empty parking space is sent to navigational panel.
The present invention also provides a kind of truck space guiding system based on image procossing and pattern-recognition, including：
The camera of multiple real-time parking stall images of collection；
Several mark rods, the mark rod is arranged on four summits or cornerwise of multiple parking stall correspondence rectangular areas
Two points；
Receiving module, the parking stall image for receiving the camera collection；
Identification module, for recognizing the mark rod in the parking stall image, according to being determined the coordinate of the mark rod
Position of the mark rod in the parking stall image, and the vehicle in the range of the mark rod is recognized, determine the number of the vehicle；
Determining module, for comparing the vehicle number and the parking stall number in the parking stall image, if of the vehicle
Number is equal to the parking stall number, it is determined that without empty parking space, if the number of the vehicle is less than the parking stall number, according to the car
Coordinate determine the position of empty parking space.
Further, the identification module, specifically for：
Judge whether the number of mark rod is less than 2, if so, then alarm signal is sent to control centre, if it is not, then judging institute
Whether the number for stating mark rod is equal to 2；
If the mark rod number is 2, judge whether two mark rods are in diagonally according to the coordinate of the mark rod
On line, if it is not, alarm signal then is sent to control centre, if, it is determined that described two mark rods are effective；
If the mark rod number is not 2, the mark rod on diagonal is chosen according to the coordinate of the mark rod.
Further, the identification module, specifically for：
Mark rod data set is obtained from picture of the database comprising mark rod；
Convolutional neural networks model is built, and the model is trained according to the data set；
The mark rod is detected using the model after training and described image to be detected, the mark rod center is stored
The coordinate of point.
Further, in addition to：
Navigational panel, the empty parking space position for receiving determining module transmission, and driven a vehicle according to the empty parking space position instruction
Direction.
The present invention uses existing monitoring camera combination mark rod, and parking stall is realized based on image procossing and pattern-recognition
Guiding, reduces the cost of system, beneficial to later maintenance, debugging and upgrading.
Brief description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing
There is the accompanying drawing used required in technology description to be briefly described, it should be apparent that, drawings in the following description are this hairs
Some bright embodiments, for those of ordinary skill in the art, without having to pay creative labor, can be with
Other accompanying drawings are obtained according to these accompanying drawings.
Fig. 1 is the parking stall bootstrap technique flow chart of the invention based on image procossing and pattern-recognition；
Fig. 2 is identification process figure of the present invention；
Fig. 3 is the truck space guiding system schematic diagram of the invention based on image procossing and pattern-recognition；
Fig. 4 is truck space guiding system another schematic diagram of the invention based on image procossing and pattern-recognition；
Fig. 5 is illustrated for the present invention based on image procossing and the truck space guiding system mark rod and camera structure of pattern-recognition
Figure.
Embodiment
To make the purpose, technical scheme and advantage of the embodiment of the present invention clearer, below in conjunction with the embodiment of the present invention
In accompanying drawing, the technical scheme in the embodiment of the present invention is clearly and completely described, it is clear that described embodiment is
A part of embodiment of the present invention, rather than whole embodiments.Based on the embodiment in the present invention, those of ordinary skill in the art
The every other embodiment obtained under the premise of creative work is not made, belongs to the scope of protection of the invention.
Fig. 1 is the parking stall bootstrap technique flow chart of the invention based on image procossing and pattern-recognition, as shown in figure 1, this reality
A method is applied, including：
Step 101, the image to be detected for receiving camera collection, described image to be detected include：Mark rod and multiple cars
Position, the mark rod is arranged on four summits or the rectangular area diagonal two of the multiple parking stall correspondence rectangular area
Summit；
Mark rod in step 102, the identification parking field picture, and the mark is determined according to the coordinate of the mark rod
Know position of the bar in the parking stall image；
Vehicle in the range of step 103, the identification mark rod, and determine the number of the vehicle；
Step 104, relatively more described vehicle number and the parking stall number in the region, if the number of the vehicle is equal to described
Parking stall number, it is determined that without empty parking space, it is true according to the coordinate of the vehicle if the number of the vehicle is less than the parking stall number
Determine the position of empty parking space.
Specifically, camera gathers the realtime graphic of multiple parking stalls, four tops of the plurality of parking stall correspondence rectangular area
Point or two summits of the rectangular area diagonal.First, those mark rods are recognized in the image of parking stall.Then, those are recognized
Vehicle in mark rod corresponding region, and parking stall number of the number of vehicle in the region prestored is compared.If knowing
Occupied in other vehicle number then region equal with the parking stall number in the region.If the vehicle number of identification is less than the area
Parking stall number in domain then determines the position of empty parking space according to the coordinate of identification vehicle.
The present invention uses existing monitoring camera combination mark rod, and parking stall is realized based on image procossing and pattern-recognition
Guiding, reduces the cost of system, beneficial to later maintenance, debugging and upgrading.
Further, the mark rod in the identification parking field picture, and being determined according to the coordinate of the mark rod
After position of the mark rod in the parking stall image, in addition to：
Judge whether the number of mark rod is less than 2, if so, then alarm signal is sent to control centre, if it is not, then judging institute
Whether the number for stating mark rod is equal to 2；
If the mark rod number is 2, judge whether two mark rods are in diagonally according to the coordinate of the mark rod
On line, if it is not, alarm signal then is sent to control centre, if, it is determined that described two mark rods are effective；
If the mark rod number is not 2, the mark rod on diagonal is chosen according to the coordinate of the mark rod.
Specifically, in actual applications, mark rod there is a situation where damage, or camera due to external force or other
Displacement caused by factor.Above-mentioned two situations once occur it is any after, can directly result in can not guide to parking stall.Cause
This, after being determined in position of the mark rod in the image of parking stall, need to further sentence to the number of mark rod in the parking stall image
It is disconnected.When the number of mark rod is less than 2, alarm signal is sent to control centre, so as to be tieed up to mark rod or camera
Repair.When the number of mark rod is equal to 2, then judge whether the position of two mark rods is located on the diagonal, not on the diagonal
Two mark rods be invalidated identification bar, it is necessary to control centre send alarm signal.Conversely, being effective mark rod.Work as mark
When dry number is not 2, then it is to have to the mark rod on diagonal to choose any of which according to the coordinate of those mark rods
Imitate mark rod.
Further, the mark rod in the parking field picture is recognized, and according to being determined the coordinate of the mark rod
Position of the mark rod in the parking stall image, including：
Mark rod data set is obtained from picture of the database comprising mark rod；
Convolutional neural networks model is built, and the model is trained according to the data set；
The mark rod is detected using the model after training and described image to be detected, the mark rod center is stored
The coordinate of point.
Specifically, the present embodiment data source is in the video recording containing marker post in some time of parking lot.
The detailed process for building convolutional neural networks model is：The convolutional neural networks model：Convolutional neural networks mould
Type comprising multiple convolutional coding structures, 5 pond layers, three full articulamentums, two ReLU layers and Dropout layers；Convolutional coding structure bag
Containing the consistent convolutional layer of three yardsticks, convolutional coding structure and pond layer alternate links, and convolutional layer top is connected to by ReLU layers,
Full articulamentum is connected after last pond layer, and prevents it from training over-fitting using Dropout layers；In convolutional layer：Each volume
The Feature Mapping figure of lamination can use multiple convolution kernels, and convolution behaviour is carried out by the Feature Mapping figure obtained to preceding layer
Make, then combination is drawn；In the layer of pond：N number of characteristic spectrum that N number of characteristic spectrum and its exported in the layer of pond is inputted is relative
Should.
Convolutional coding structure includes 5 layers of different scale in the present embodiment, and every layer of convolutional coding structure will carry out three subdimension identicals
Convolution, convolution kernel is 3 × 3.
First layer convolutional coding structure z1=W1*I+B1, wherein I is input picture, and size is 224 × 224, W1Represent f1×f1×
n1Wave filter, convolution kernel size f1=3, step-length is 1, and convolution kernel number is n1=64, that is, export n1Individual characteristic pattern, * represents volume
Product operation, B1Represent n1Tie up deviation.z1Represent the characteristic pattern exported after this layer.224 × 224 image passes through this convolution, generation
The characteristic pattern that 64 sizes are 110 × 110.
There is one layer of pond behind first layer convolutional coding structurez1Represent the feature inputted after this layer
Figure, convolution kernel size fp=3, stride represent step-length, and step-length is 2.There is one layer of Maxout activation behind pond
I is this node layer, and j is the implicit node corresponding to this layer of each node.J span is [1, k], and k is maxout layers of institute
The parameter needed.p1Represent the characteristic pattern of this layer of input, F1The characteristic pattern of this layer of output is represented, final output characteristic pattern number isIndividual 110 × 110 blurred picture passes through this pond, the characteristic pattern that 128 sizes of generation are 55 × 55.
Second layer convolutional coding structure z2=W2*F1+B2, wherein W2RepresentWave filter, convolution kernel size f2=
3, convolution kernel number is n2=128, step-length is 1, that is, exports n2Individual characteristic pattern, * represents convolution operation, B2Represent n2Tie up deviation.
F1Represent this layer of input feature vector figure, z2Represent the characteristic pattern exported after this layer.55 × 55 characteristic patterns pass through this convolution, generation
The characteristic pattern that 128 sizes are 26 × 26.
There is one layer of pond behind second layer convolutional coding structurez2The characteristic pattern inputted after this layer is represented,
Convolution kernel size fp=3, stride represent step-length, and step-length is 2.There is one layer of Maxout activation behind pondi
For this node layer, j is the implicit node corresponding to this layer of each node.J span is [1, k], and k is for needed for maxout layers
The parameter wanted.p2Represent the characteristic pattern of this layer of input, F2The characteristic pattern of this layer of output is represented, final output characteristic pattern number is
Individual 26 × 26 image passes through this pond, the characteristic pattern that 256 sizes of generation are 13 × 13.
Third layer convolutional coding structure z3=W3*F2+B3, wherein W3RepresentWave filter, convolution kernel size f3
=3, convolution kernel number is n3=256, step-length is 1, that is, exports n3Individual characteristic pattern, * represents convolution operation, B3Represent n3Dimension is inclined
Difference.F2Represent this layer of input feature vector figure, z3Represent the characteristic pattern exported after this layer.13 × 13 characteristic patterns pass through this convolution, raw
The characteristic pattern for being 13 × 13 into 256 sizes.
There is one layer of pond behind third layer convolutional coding structurez3The characteristic pattern inputted after this layer is represented,
Convolution kernel size fp=3, stride represent step-length, and step-length is 2.There is one layer of Maxout activation behind pondi
For this node layer, j is the implicit node corresponding to this layer of each node.J span is [1, k], and k is for needed for maxout layers
The parameter wanted.p3Represent the characteristic pattern of this layer of input, F3The characteristic pattern of this layer of output is represented, final output characteristic pattern number is
Individual 13 × 13 image passes through this pond, the characteristic pattern that 512 sizes of generation are 6 × 6.
4th layer of convolutional coding structure z4=W4*F3+B4, wherein W4RepresentWave filter, convolution kernel size f4
=3, convolution kernel number is n4=512, step-length is 1, that is, exports n4Individual characteristic pattern, * represents convolution operation, B4Represent n4Dimension is inclined
Difference.F3Represent this layer of input feature vector figure, z4Represent the characteristic pattern exported after this layer.6 × 6 characteristic patterns pass through this convolution, generation
The characteristic pattern that 512 sizes are 6 × 6.
There is one layer of pond behind the 4th layer of convolutional coding structurez4The characteristic pattern inputted after this layer is represented,
Convolution kernel size fp=3, stride represent step-length, and step-length is 2.There is one layer of Maxout activation behind pondi
For this node layer, j is the implicit node corresponding to this layer of each node.J span is [1, k], and k is for needed for maxout layers
The parameter wanted.p4Represent the characteristic pattern of this layer of input, F4The characteristic pattern of this layer of output is represented, final output characteristic pattern number is
Individual 6 × 6 image passes through this pond, the characteristic pattern that 512 sizes of generation are 3 × 3.
Layer 5 convolutional coding structure z5=W5*F4+B5, wherein W5RepresentWave filter, convolution kernel size f5
=3, convolution kernel number is n5=512, step-length is 1, that is, exports n5Individual characteristic pattern, * represents convolution operation, B5Represent n5Dimension is inclined
Difference.F4Represent this layer of input feature vector figure, z5Represent the characteristic pattern exported after this layer.3 × 3 characteristic patterns pass through this convolution, generation
The characteristic pattern that 512 sizes are 3 × 3.
After the completion of last layer of convolutional coding structure processing, then three full articulamentums are linked into, 2 sizes of generation are 4096 × 1
1 size of characteristic vector and generation be 1000 × 1 characteristic vector, these three full articulamentums are respectively used to classification and frame is returned
Return.Contain 2 elements in frame returns layer, export to represent be the probability of target and be not target probability, side
What full articulamentum was finally exported in frame recurrence layer is target area position.The target area position is the frame coordinate of marker post.
After initialization, training is iterated to the convolutional neural networks model of structure using stochastic gradient descent method, often changed
For the subgradient of one-time detection one and the value of loss function, to obtain in network architecture each weighted value W and bias b most
Excellent solution, obtains the optimal convolutional neural networks model of this training after iteration is multiple.
Further, the vehicle in the range of the identification mark rod, including：
The central point for connecting the mark rod determines the scope of parking stall measure；
Vehicle data collection is obtained from picture of the database comprising vehicle, convolutional neural networks model is built, and according to institute
State data set and train the model；
The vehicle is detected using image to be detected in the range of the model and parking stall measure after training.
Specifically, the present embodiment data set derives from the Cars data sets that Stanford University creates, 196 class vehicle figures
Piece, totally 16185 pictures.The CompCars data sets that Tang Xiao gulls team of Hong Kong Chinese University sets up, including 136727 vehicles
Vehicle pictures and the local picture of 27618 vehicles.
The detailed process for building convolutional neural networks model is：Specifically, the convolutional neural networks model：Convolution god
Through network model comprising multiple convolutional layers, 3 pond layers, two full articulamentums, two ReLU layers and Dropout layers；Convolution
Layer and pond layer alternate links, and convolutional layer top is connected to by ReLU layers, full articulamentum is connected after last pond layer,
And prevent it from training over-fitting using Dropout layers；In convolutional layer：The Feature Mapping figure of each convolutional layer can use multiple
Convolution kernel, convolution operation is carried out by the Feature Mapping figure obtained to preceding layer, and then combination is drawn；In the layer of pond：Pond layer
N number of characteristic spectrum that N number of characteristic spectrum and its of middle output are inputted is corresponding.
Convolutional layer includes 6 layers of different scale in the present embodiment, and convolution kernel is respectively 7 × 7,5 × 5,3 × 3,3 × 3,3 ×
3,3 × 3, active coating includes 3 layers, after the every layer pond.
First layer convolution z1=W1*I+B1, wherein I is input picture, and size is 224 × 224, W1Represent f1×f1×n1Filter
Ripple device, convolution kernel size f1=7, step-length is 2, and convolution kernel number is n1=96, that is, export n1Individual characteristic pattern, * represents that convolution is grasped
Make, B1Represent n1Tie up deviation.z1Represent the characteristic pattern exported after this layer.224 × 224 image passes through this convolution, generates 96
Size is 110 × 110 characteristic pattern.
There is one layer of pond behind first layer convolutionz1The characteristic pattern inputted after this layer is represented, is rolled up
Product core size fp=3, stride represent step-length, and step-length is 2.There is one layer of Maxout activation behind pondI is
This node layer, j is the implicit node corresponding to this layer of each node.J span is [1, k], and k is maxout layers required
Parameter.p1Represent the characteristic pattern of this layer of input, F1The characteristic pattern of this layer of output is represented, final output characteristic pattern number isIt is individual
110 × 110 blurred picture passes through this pond, the characteristic pattern that 96 sizes of generation are 55 × 55.
Second layer convolution z2=W2*F1+B2, wherein W2RepresentWave filter, convolution kernel size f2=5, volume
Product core number is n2=256, step-length is 2, that is, exports n2Individual characteristic pattern, * represents convolution operation, B2Represent n2Tie up deviation.F1Table
Show this layer of input feature vector figure, z2Represent the characteristic pattern exported after this layer.55 × 55 characteristic patterns pass through this convolution, generation 256
Individual size is 26 × 26 characteristic pattern.
There is one layer of pond behind second layer convolutionz2The characteristic pattern inputted after this layer is represented,
Convolution kernel size fp=3, stride represent step-length, and step-length is 2.There is one layer of Maxout activation behind pond
I is this node layer, and j is the implicit node corresponding to this layer of each node.J span is [1, k], and k is maxout layers of institute
The parameter needed.p2Represent the characteristic pattern of this layer of input, F2The characteristic pattern of this layer of output is represented, final output characteristic pattern number isIndividual 26 × 26 image passes through this pond, the characteristic pattern that 256 sizes of generation are 13 × 13.
Third layer convolution z3=W3*F2+B3, wherein W3RepresentWave filter, convolution kernel size f3=3,
Convolution kernel number is n3=384, step-length is 1, that is, exports n3Individual characteristic pattern, * represents convolution operation, B3Represent n3Tie up deviation.F2
Represent this layer of input feature vector figure, z3Represent the characteristic pattern exported after this layer.13 × 13 characteristic patterns pass through this convolution, generation
The characteristic pattern that 384 sizes are 13 × 13.
4th layer of convolution z4=W4*F3+B4, wherein W4RepresentWave filter, convolution kernel size f4=3,
Convolution kernel number is n4=384, step-length is 1, that is, exports n4Individual characteristic pattern, * represents convolution operation, B4Represent n4Tie up deviation.F3
Represent this layer of input feature vector figure, z4Represent the characteristic pattern exported after this layer.13 × 13 characteristic patterns pass through this convolution, generation
The characteristic pattern that 384 sizes are 13 × 13.
Layer 5 convolution z5=W5*F4+B5, wherein W5RepresentWave filter, convolution kernel size f5=3,
Convolution kernel number is n5=384, step-length is 1, that is, exports n5Individual characteristic pattern, * represents convolution operation, B5Represent n5Tie up deviation.F4
Represent this layer of input feature vector figure, z5Represent the characteristic pattern exported after this layer.13 × 13 characteristic patterns pass through this convolution, generation
The characteristic pattern that 384 sizes are 13 × 13.
Layer 6 convolution z6=W6*F5+B6, wherein W5RepresentWave filter, convolution kernel size f6=3,
Convolution kernel number is n6=256, step-length is 1, that is, exports n6Individual characteristic pattern, * represents convolution operation, B6Represent n6Tie up deviation.F5
Represent this layer of input feature vector figure, z6Represent the characteristic pattern exported after this layer.13 × 13 characteristic patterns pass through this convolution, generation
The characteristic pattern that 256 sizes are 13 × 13.
There is one layer of pond behind layer 6 convolutionz2The characteristic pattern inputted after this layer is represented,
Convolution kernel size fp=3, stride represent step-length, and step-length is 2.There is one layer of Maxout activation behind pond
I is this node layer, and j is the implicit node corresponding to this layer of each node.J span is [1, k], and k is maxout layers of institute
The parameter needed.p3Represent the characteristic pattern of this layer of input, F3The characteristic pattern of this layer of output is represented, final output characteristic pattern number isIndividual 26 × 26 image passes through this pond, the characteristic pattern that 256 sizes of generation are 6 × 6.
After the completion of last layer of pond layer processing, then two full articulamentums are linked into, 1 size of generation is 4096 × 1
Characteristic vector, the two full articulamentums are respectively used to classification and frame is returned.2 elements are contained in frame returns layer, it is defeated
Go out to represent be the probability of target and be not target probability, what frame returned that full articulamentum in layer finally exports is target
Regional location.
After initialization, training is iterated to the convolutional neural networks model of structure using stochastic gradient descent method, often changed
For the subgradient of one-time detection one and the value of loss function, to obtain in network architecture each weighted value W and bias b most
Excellent solution, obtains the optimal convolutional neural networks model of this training after iteration is multiple.
Further, it is described, the position of empty parking space is determined according to the coordinate of the vehicle, including：
The corresponding rectangle of the mark rod is determined according to the coordinate of two mark rods on the diagonal, according to the square
Parking stall number in shape divides equally the rectangle for several parking stall rectangles；
The position of empty parking space is determined according to the coordinate of the vehicle and the parking stall rectangle.
Specifically, if four point coordinates that marker post surrounds rectangle are
Parking stall number is n in rectangle, then the coordinate of the parking stall rectangle marked off should be (ax1, ay1)(ax1, by1)(ax2,
ay2)(ax2, by2) (axn, ayn)(axn, byn)(bx1, ay1)(bx1, by1)(bx2, ay2)(bx2, by2)…(bxn, ayn)
(bxn, yn)
If detecting, vehicle number is less than parking stall number, and the coordinate with vehicle is sought according to testing result
The coordinate of immediate parking stall, i.e.,
Using nearest parking stall as the parking stall of the vehicle detected, mark is designated as n, lack mark is designated as n-th of sky
Parking stall.
Further, after the coordinate according to the vehicle and parking stall rectangle determine the position of empty parking space, go back
Including：
The position of the empty parking space is sent to navigational panel.
Specifically, navigational panel is shown in a panel region nearby with the presence or absence of the navigational panel correspondence of empty parking space, i.e., one
Parking stall situation under multiple cameras, label is done by camera, and camera tool can be corresponded to out according to the parking stall map in parking lot
Which subregion of body, if there is empty parking space in this region, shows the quantity of empty parking space, and it is indicated by an arrow go out the navigational panel taken the photograph with this
As the position relation of head.
Fig. 3 is the truck space guiding system schematic diagram of the invention based on image procossing and pattern-recognition, as shown in figure 3, this reality
A system is applied, including：
The camera 101 of multiple real-time parking stall images of collection；
Several mark rods 102, the mark rod is arranged on four summits or diagonal of multiple parking stall correspondence rectangular areas
Two points of line；
Receiving module 103, the parking stall image for receiving the camera collection；
Identification module 104, for recognizing the mark rod in the parking stall image, institute is determined according to the coordinate of the mark rod
Position of the mark rod in the parking stall image is stated, and recognizes the vehicle in the range of the mark rod, of the vehicle is determined
Number；
Determining module 105, for comparing the vehicle number and the parking stall number in the parking stall image, if the vehicle
Number is equal to the parking stall number, it is determined that without empty parking space, if the number of the vehicle is less than the parking stall number, according to described
The coordinate of vehicle determines the position of empty parking space.
Specifically, parking lot of the invention can be for parking garage or open parking ground, and multiple cameras can
There are different mounting means, the camera in parking lot can directly select existing monitoring camera in parking lot indoors, its
It is required that needing to ensure without monitoring dead angle.Two kinds of mounting means can be had according to the difference of environment in open parking ground, building is depended on
Install or rod-type is installed, the vehicle of detection is needed in the highly desirable guarantee visual angle of installation blocking for larger area should not occurs.
As shown in figure 5, corresponding to the mark rod of setting within sweep of the eye 102 of each camera 101, the mark rod is used for school
The identification region of the accurate camera.The diagonal for the corresponding rectangular area in multiple parking stalls that can be recognized in the camera is set
Two mark rods are put, four summits of the corresponding rectangular area in multiple parking stalls that can also be recognized in the camera set 4 marks
Know bar.Mark rod set position to ensure camera within sweep of the eye and remain to photograph in camera weak vibrations
Mark rod, mark rod can use the parking stall number in the range of the alternate grade of reddish yellow more strikingly color, mark rod to be integer.It is indoor
In parking lot mark rod with outdoor mark rod similarly.Multiple cameras will enter line label, and multiple camera views can have repetition
Region, but to ensure each parking stall at least in the corresponding mark rod region of a camera.
Further, the identification module, specifically for：
Judge whether the number of mark rod is less than 2, if so, then alarm signal is sent to control centre, if it is not, then judging institute
Whether the number for stating mark rod is equal to 2；
If the mark rod number is 2, judge whether two mark rods are in diagonally according to the coordinate of the mark rod
On line, if it is not, alarm signal then is sent to control centre, if, it is determined that described two mark rods are effective；
If the mark rod number is not 2, the mark rod on diagonal is chosen according to the coordinate of the mark rod.
Further, the identification module, specifically for：
Mark rod data set is obtained from picture of the database comprising mark rod；
Convolutional neural networks model is built, and the model is trained according to the data set；
The mark rod is detected using the model after training and described image to be detected, the mark rod center is stored
The coordinate of point.
Further, as shown in figure 4, the system of the present embodiment, in addition to：
Navigational panel 106, the empty parking space position for receiving determining module transmission, and according to the empty parking space position instruction row
Car direction.
Present system is identical with the former topic of the method shown in Fig. 1, and here is omitted.
Compared with prior art, the present invention has following beneficial effect：
(1) inexpensive multiple spot parking stall data snooping
Multiple spot parking stall data snooping shoots the parking stall occupancy situation in parking lot, the position set according to camera using camera
Put, the situation of multiple parking stalls can be shot simultaneously, compared to single-point parking stall data snooping, the benefit of multiple spot parking stall data snooping exists
In cost can be reduced at double.
(2) real-time parking stall data snooping
Parking stall measure data of the present invention are accurate, real-time is high.At present through experiment, the time loss of a frame high-definition image is handled
In 0.04s or so, then under conditions of serial, the data of at least 20 cameras can be still handled in 1s, it is parallel under the conditions of when
The data of more multi-cam can be so handled, therefore the real-time of detection can be ensured, meanwhile, install and meet in camera position
It is required that on the premise of, the accuracy rate that multiple target is detected simultaneously can reach more than 99.5%.
(3) it can be used for outdoor detection
The scene that prior art is adapted to is generally indoor, when applying in outdoor environment, due to there is light, the weather such as wet weather
The influence of situation, can greatly cut down accuracy rate, and the present invention can not only realize the parking stall guiding of parking garage, can also realize
The accurate detection of outdoor conditions parking space.
Finally it should be noted that：Various embodiments above is merely illustrative of the technical solution of the present invention, rather than its limitations；To the greatest extent
The present invention is described in detail with reference to foregoing embodiments for pipe, it will be understood by those within the art that：Its according to
The technical scheme described in foregoing embodiments can so be modified, or which part or all technical characteristic are entered
Row equivalent substitution；And these modifications or replacement, the essence of appropriate technical solution is departed from various embodiments of the present invention technology
The scope of scheme.
Claims (10)
1. a kind of parking stall bootstrap technique based on image procossing and pattern-recognition, it is characterised in that including：
Image to be detected of camera collection is received, described image to be detected includes：Mark rod and multiple parking stalls, the mark rod
It is arranged on four summits of the multiple parking stall correspondence rectangular area or cornerwise two points；
Mark rod in the identification parking field picture, and determine the mark rod in the car according to the coordinate of the mark rod
Position in bit image；
The vehicle in the range of the mark rod is recognized, and determines the number of the vehicle；
Compare the vehicle number and the parking stall number in the scope, if the number of the vehicle is equal to the parking stall number, really
Fixed no empty parking space, if the number of the vehicle is less than the parking stall number, empty parking space is determined according to the coordinate of the vehicle
Position.
2. according to the method described in claim 1, it is characterised in that the mark rod in the identification parking field picture, and
After position of the mark rod in the parking stall image being determined according to the coordinate of the mark rod, in addition to：
Judge whether the number of mark rod is less than 2, if so, then alarm signal is sent to control centre, if it is not, then judging the mark
Whether the number for knowing bar is equal to 2；
If the mark rod number is 2, judge whether two mark rods are located on the diagonal according to the coordinate of the mark rod,
If it is not, alarm signal then is sent to control centre, if, it is determined that described two mark rods are effective；
If the mark rod number is not 2, the mark rod on diagonal is chosen according to the coordinate of the mark rod.
3. method according to claim 1 or 2, it is characterised in that the mark rod in the identification parking field picture, and root
Position of the mark rod in the parking stall image is determined according to the coordinate of the mark rod, including：
Mark rod data set is obtained from picture of the database comprising mark rod；
Convolutional neural networks model is built, and the model is trained according to the data set；
The mark rod is detected using the model after training and described image to be detected, the mark rod central point is stored
Coordinate.
4. method according to claim 3, it is characterised in that the vehicle in the range of the identification mark rod, including：
The central point for connecting the mark rod determines the scope of parking stall measure；
Vehicle data collection is obtained from picture of the database comprising vehicle, convolutional neural networks model is built, and according to the number
The model is trained according to collection；
The vehicle is detected using image to be detected in the range of the model and parking stall measure after training.
5. method according to claim 2, it is characterised in that described, empty parking space is determined according to the coordinate of the vehicle
Position, including：
The corresponding rectangle of the mark rod is determined according to the coordinate of two mark rods on the diagonal, according in the rectangle
Parking stall number divide equally the rectangle for several parking stall rectangles；
The position of empty parking space is determined according to the coordinate of the vehicle and the parking stall rectangle.
6. method according to claim 5, it is characterised in that the coordinate according to the vehicle and parking stall rectangle
After the position for determining empty parking space, in addition to：
The position of the empty parking space is sent to navigational panel.
7. a kind of truck space guiding system based on image procossing and pattern-recognition, it is characterised in that including：
The camera of multiple real-time parking stall images of collection；
Several mark rods, the mark rod is arranged on four summits or cornerwise two of multiple parking stall correspondence rectangular areas
Point；
Receiving module, the parking stall image for receiving the camera collection；
Identification module, for recognizing the mark rod in the parking stall image, the mark is determined according to the coordinate of the mark rod
Position of the bar in the parking stall image, and the vehicle in the range of the mark rod is recognized, determine the number of the vehicle；
Determining module, for comparing the vehicle number and the parking stall number in the parking stall image, if number of the vehicle etc.
In the parking stall number, it is determined that without empty parking space, if the number of the vehicle is less than the parking stall number, according to the vehicle
Coordinate determines the position of empty parking space.
8. system according to claim 7, it is characterised in that the identification module, specifically for：
Judge whether the number of mark rod is less than 2, if so, then alarm signal is sent to control centre, if it is not, then judging the mark
Whether the number for knowing bar is equal to 2；
If the mark rod number is 2, judge whether two mark rods are located on the diagonal according to the coordinate of the mark rod,
If it is not, alarm signal then is sent to control centre, if, it is determined that described two mark rods are effective；
If the mark rod number is not 2, the mark rod on diagonal is chosen according to the coordinate of the mark rod.
9. system according to claim 8, it is characterised in that the identification module, specifically for：
Mark rod data set is obtained from picture of the database comprising mark rod；
Convolutional neural networks model is built, and the model is trained according to the data set；
The mark rod is detected using the model after training and described image to be detected, the mark rod central point is stored
Coordinate.
10. system according to claim 7, it is characterised in that also include：
Navigational panel, the empty parking space position for receiving determining module transmission, and according to the empty parking space position instruction direction of traffic.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710447805.4A CN107067813A (en) | 2017-06-14 | 2017-06-14 | A kind of parking stall bootstrap technique and system based on image procossing and pattern-recognition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710447805.4A CN107067813A (en) | 2017-06-14 | 2017-06-14 | A kind of parking stall bootstrap technique and system based on image procossing and pattern-recognition |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107067813A true CN107067813A (en) | 2017-08-18 |
Family
ID=59595279
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710447805.4A Pending CN107067813A (en) | 2017-06-14 | 2017-06-14 | A kind of parking stall bootstrap technique and system based on image procossing and pattern-recognition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107067813A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107665603A (en) * | 2017-09-06 | 2018-02-06 | 哈尔滨工程大学 | A kind of real-time detection method for judging parking stall and taking |
| CN108108689A (en) * | 2017-12-18 | 2018-06-01 | 北京工业大学 | A kind of intelligent parking method for detecting parking stalls |
| CN109086708A (en) * | 2018-07-25 | 2018-12-25 | 深圳大学 | A kind of parking space detection method and system based on deep learning |
| CN109147382A (en) * | 2018-09-25 | 2019-01-04 | 深圳市元征科技股份有限公司 | A kind of intelligent parking method and system |
| CN110097776A (en) * | 2018-01-30 | 2019-08-06 | 杭州海康威视数字技术股份有限公司 | A kind of method for detecting parking stalls, monitor camera and monitor terminal |
| US20200394745A1 (en) * | 2019-06-14 | 2020-12-17 | Toyota Motor North America, Inc. | Parking monitoring and assistance for transports |
| US11626019B2 (en) | 2019-06-14 | 2023-04-11 | Toyota Motor North America, Inc. | Parking monitoring and assistance for transports |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101656023A (en) * | 2009-08-26 | 2010-02-24 | 西安理工大学 | Management method of indoor car park in video monitor mode |
| CN101894482A (en) * | 2010-07-15 | 2010-11-24 | 东南大学 | Video technology-based roadside vacant parking position wireless network detection system and method |
| CN104282169A (en) * | 2014-09-29 | 2015-01-14 | 四川川大智胜软件股份有限公司 | Embedded integration parking space detector and detecting method thereof |
| JP5988185B1 (en) * | 2016-01-19 | 2016-09-07 | 株式会社大都製作所 | Parking lot management system |
| CN106652551A (en) * | 2016-12-16 | 2017-05-10 | 浙江宇视科技有限公司 | Parking stall detection method and device |
-
2017
- 2017-06-14 CN CN201710447805.4A patent/CN107067813A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101656023A (en) * | 2009-08-26 | 2010-02-24 | 西安理工大学 | Management method of indoor car park in video monitor mode |
| CN101894482A (en) * | 2010-07-15 | 2010-11-24 | 东南大学 | Video technology-based roadside vacant parking position wireless network detection system and method |
| CN104282169A (en) * | 2014-09-29 | 2015-01-14 | 四川川大智胜软件股份有限公司 | Embedded integration parking space detector and detecting method thereof |
| JP5988185B1 (en) * | 2016-01-19 | 2016-09-07 | 株式会社大都製作所 | Parking lot management system |
| CN106652551A (en) * | 2016-12-16 | 2017-05-10 | 浙江宇视科技有限公司 | Parking stall detection method and device |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107665603A (en) * | 2017-09-06 | 2018-02-06 | 哈尔滨工程大学 | A kind of real-time detection method for judging parking stall and taking |
| CN107665603B (en) * | 2017-09-06 | 2020-06-16 | 哈尔滨工程大学 | Real-time detection method for judging parking space occupation |
| CN108108689A (en) * | 2017-12-18 | 2018-06-01 | 北京工业大学 | A kind of intelligent parking method for detecting parking stalls |
| CN110097776A (en) * | 2018-01-30 | 2019-08-06 | 杭州海康威视数字技术股份有限公司 | A kind of method for detecting parking stalls, monitor camera and monitor terminal |
| CN110097776B (en) * | 2018-01-30 | 2021-05-14 | 杭州海康威视数字技术股份有限公司 | Parking space detection method, monitoring camera and monitoring terminal |
| CN109086708A (en) * | 2018-07-25 | 2018-12-25 | 深圳大学 | A kind of parking space detection method and system based on deep learning |
| CN109147382A (en) * | 2018-09-25 | 2019-01-04 | 深圳市元征科技股份有限公司 | A kind of intelligent parking method and system |
| CN109147382B (en) * | 2018-09-25 | 2022-01-11 | 深圳市元征科技股份有限公司 | Intelligent parking method and system |
| US20200394745A1 (en) * | 2019-06-14 | 2020-12-17 | Toyota Motor North America, Inc. | Parking monitoring and assistance for transports |
| US11514544B2 (en) * | 2019-06-14 | 2022-11-29 | Toyota Motor North America, Inc. | Parking monitoring and assistance for transports |
| US11626019B2 (en) | 2019-06-14 | 2023-04-11 | Toyota Motor North America, Inc. | Parking monitoring and assistance for transports |
| US11948460B2 (en) | 2019-06-14 | 2024-04-02 | Toyota Motor North America, Inc. | Parking monitoring and assistance for transports |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107067813A (en) | A kind of parking stall bootstrap technique and system based on image procossing and pattern-recognition | |
| Ali et al. | Real-time multiple damage mapping using autonomous UAV and deep faster region-based neural networks for GPS-denied structures | |
| CN108961790B (en) | Bad weather early warning management system and method based on four-dimensional live-action traffic simulation | |
| CN106023605B (en) | A kind of method for controlling traffic signal lights based on depth convolutional neural networks | |
| DE102014223363B4 (en) | Method and device for localizing a motor vehicle in a fixed reference map | |
| Pavlidis et al. | Urban surveillance systems: from the laboratory to the commercial world | |
| CN111553201B (en) | Traffic light detection method based on YOLOv3 optimization algorithm | |
| CN106097755A (en) | For identifying the method parked place and/or vacate place | |
| CN106980657A (en) | A kind of track level electronic map construction method based on information fusion | |
| CN111339905B (en) | CIM well lid state visual detection system based on deep learning and multiple visual angles | |
| CN109255286A (en) | A kind of quick detection recognition method of unmanned plane optics based on YOLO deep learning network frame | |
| CN111402632B (en) | Risk prediction method for pedestrian movement track at intersection | |
| CN111723704A (en) | Raspberry pie-based van body door opening monitoring method | |
| CN109101932A (en) | The deep learning algorithm of multitask and proximity information fusion based on target detection | |
| CN106845359A (en) | Tunnel portal driving prompt apparatus and method based on infrared emission | |
| CN108648210B (en) | Rapid multi-target detection method and device under static complex scene | |
| CN110796360A (en) | Fixed traffic detection source multi-scale data fusion method | |
| CN110083099A (en) | One kind meeting automobile function safety standard automatic Pilot architecture system and working method | |
| CN114898319A (en) | Vehicle type recognition method and system based on multi-sensor decision-level information fusion | |
| CN111951598B (en) | Vehicle tracking monitoring method, device and system | |
| CN115292913A (en) | Vehicle-road-cooperation-oriented drive test perception simulation system | |
| Muller | Drivetruth: Automated autonomous driving dataset generation for security applications | |
| CN114298163A (en) | Online road condition detection system and method based on multi-source information fusion | |
| CN113343933A (en) | Airport scene monitoring method based on video target identification and positioning | |
| CN111316324A (en) | Automatic driving simulation system, method, equipment and storage medium |
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 |
Application publication date: 20170818 |
|
| RJ01 | Rejection of invention patent application after publication |