CN106971544B - A kind of direct method that vehicle congestion is detected using still image - Google Patents
A kind of direct method that vehicle congestion is detected using still image Download PDFInfo
- Publication number
- CN106971544B CN106971544B CN201710339106.8A CN201710339106A CN106971544B CN 106971544 B CN106971544 B CN 106971544B CN 201710339106 A CN201710339106 A CN 201710339106A CN 106971544 B CN106971544 B CN 106971544B
- Authority
- CN
- China
- Prior art keywords
- vehicle
- image
- information
- convolutional neural
- neural networks
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0125—Traffic data processing
- G08G1/0133—Traffic data processing for classifying traffic situation
Abstract
The invention discloses a kind of methods that vehicle congestion is directly detected using still image comprising step: data prediction;The training of multitask convolutional neural networks;Detection image;Vehicle queue length and congestion determine.Image aspects, vehicle number and vehicle closeness are determined by data prediction.Gray image and illumination invariant image after sample image is converted by original RGB image, as the input of multitask convolutional neural networks, Lai Xunlian multitask convolutional neural networks are trained image aspects, vehicle number and vehicle closeness parallel.By the multitask convolutional neural networks after image to be detected input training, image aspects corresponding with image to be detected, vehicle number and vehicle closeness or combinations thereof information are obtained.According to image aspects, vehicle number and vehicle closeness comprehensive descision vehicle congestion situation.The present invention does not need identification vehicle moving process, and the judgement that number of vehicles realizes congestion detection directly can be obtained using still image.
Description
Technical field
The present invention relates to a kind of vehicle congestion detection methods, and in particular to a kind of directly to detect vehicle using still image
The method of congestion.
Background technique
With the increase of urban population and urban traffic flow, city, which is especially metropolitan traffic problems commonly, becomes burnt
Point problem.The congested in traffic blockage problem of urban road, which has become, to be restricted economic development, reduces people's living standard, weakens economy
One of bottleneck of vigor.So how quickly and efficiently detection congestion in road is the needs for complying with social development.
Traffic congestion, for traveler, feeling mainly to time and speed, i.e., vehicle is in road or intersection
Upper queuing is slowly mobile.Definition is set forth to intersection and congested link in the Ministry of Public Security, China: vehicle is in no signal
It is obstructed on driveway at the intersection of lamp control and queue length is more than the crossroad of 250 meters or vehicle in Signalized control
Mouthful, No. 3 green lights, which are shown, is not defined as intersection by the state at crossing;Congested link be then defined as vehicle on driveway by
Resistance and queue length are more than 1 kilometer of state.
From the prior art, a traditional methods are started with from vehicle itself, are detected, are utilized using sensor
GPS positioning technology carries out congestion detection, and in addition there are the means such as ground induction coil detection, Data mining, microwave detection.But
All there is some shortcomingss in these traditional detection methods, such as the cost of investment detected is high, physical equipment holds
It is easy to damage, be difficult to the disadvantages of safeguarding.
Another kind of is to detect traffic congestion using video image technology, using video image processing technology, in video
Vehicle is identified and is tracked, and combines some speed, azimuth information, is detected to congestion.Such video detection technology
There is many disadvantages, the mutual serious shielding of vehicle, the track following based on single unit vehicle tends to fail.In addition,
The height of traffic camera shooting, angle is also different, some vehicles are very fuzzy very in video or image, this also gives detection band
Carry out extreme difficulties, causes the inaccuracy of result.
Summary of the invention
The present invention proposes a kind of method that vehicle congestion is directly detected using still image, and it is mobile not need identification vehicle
The judgement that number of vehicles realizes congestion detection can be obtained in process.
Solution of the invention is: a method of vehicle congestion directly being detected using still image, suitable for counting
It calculates and is executed in machine equipment comprising following steps:
Data prediction: width sample image every in training sample image data is divided into m × n subgraph, takes people
The form of work calibration is that each subgraph assigns its two attribute information: vehicle number information and vehicle closeness information;Original image
Image aspects information is assigned according to camera shooting grease head highness visual angle;
The training of multitask convolutional neural networks: multitask convolutional neural networks are multilayer convolution, and sample image is passed through original
Gray image and illumination invariant image after the conversion of beginning RGB image are more to train as the input of multitask convolutional neural networks
Task convolutional neural networks are trained image aspects information, vehicle number information and vehicle closeness information parallel;
Detection image: the multitask convolutional neural networks after image to be detected input training obtain and image to be detected
Corresponding image aspects information, vehicle number information and vehicle closeness information or combinations thereof information;
Vehicle queue length and congestion determine: comprehensive according to image aspects information, vehicle number information and vehicle closeness information
Conjunction judges vehicle congestion situation.
As a further improvement of the foregoing solution, vehicle closeness information is determined by vehicle number information.
Further, vehicle closeness information defines 4 ranks: the corresponding vehicle of vehicle number information of 0-5 vehicle is intensive
Spending information is 0;The corresponding vehicle closeness information of the vehicle number information of 6-10 vehicle is 1;The vehicle number information pair of 11-15 vehicle
The vehicle closeness information answered is 2;The corresponding vehicle closeness information of the vehicle number information of 16 vehicles is 3.
As a further improvement of the foregoing solution, image aspects information by road camera mounting height and shooting visual angle
It determines.
Further, image aspects information is divided into basic, normal, high by three according to the mounting height and shooting visual angle of road camera
Kind visual angle.
As a further improvement of the foregoing solution, multitask convolutional neural networks carry out convolution a: side to both direction
To for image aspects information, other direction is vehicle number information and vehicle closeness information.
Further, loss function is defined in multitask convolutional neural networks:
One, vehicle count loss function: vehicle count is regarded as depth regression problem, input is whole image, and output is
The vehicle number of different subgraphs minimizes Euclidean lossUsing formulaWherein, k is the rope of subgraph
Draw, Gk and Pk are the true of kth subgraph and prediction vehicle number;
Two, vehicle closeness loss function: vehicle closeness is defined according to vehicle number, the loss of vehicle closenessIt adopts
Use formulaWherein k is the index of subgraph, and pkm is the areal densities grade of kth subgraph;
Three, image aspects loss function: image aspects lossUsing
Wherein, C is the sum of image aspects classification, yic
And picIt is prediction and true picture visual angle respectively;
By above three loss function, the multitask convolutional neural networks after training obtain convolutional neural networks model:Wherein, N is the quantity of training sample image, and λ is for balancing different loss letters
Number, W and b are model parameters, and the multitask convolutional neural networks that λ is set as after 0 training will not consider image aspects.
As a further improvement of the foregoing solution, multitask convolutional neural networks share 3 convolutional layers.
The present invention also provides a kind of storage equipment, wherein storing a plurality of instruction, described instruction is suitable for processor and loads and hold
Row, a plurality of instruction are as follows:
Data prediction: width sample image every in training sample image data is divided into m × n subgraph, takes people
The form of work calibration is that each subgraph assigns its two attribute information: vehicle number information and vehicle closeness information;Original image
Image aspects information is assigned according to camera shooting grease head highness visual angle;
The training of multitask convolutional neural networks: multitask convolutional neural networks are multilayer convolution, and sample image is passed through original
Gray image and illumination invariant image after the conversion of beginning RGB image are more to train as the input of multitask convolutional neural networks
Task convolutional neural networks are trained image aspects information, vehicle number information and vehicle closeness information parallel;
Detection image: the multitask convolutional neural networks after image to be detected input training obtain and image to be detected
Corresponding image aspects information, vehicle number information and vehicle closeness information or combinations thereof information;
Vehicle queue length and congestion determine: comprehensive according to image aspects information, vehicle number information and vehicle closeness information
Conjunction judges vehicle congestion situation.
The present invention also provides a kind of mobile communication terminals comprising:
Processor is adapted for carrying out various instructions;And
Equipment is stored, is suitable for storing a plurality of instruction, described instruction is suitable for processor and loads and execute;
The a plurality of instruction are as follows:
Data prediction: width sample image every in training sample image data is divided into m × n subgraph, takes people
The form of work calibration is that each subgraph assigns its two attribute information: vehicle number information and vehicle closeness information;Original image
Image aspects information is assigned according to camera shooting grease head highness visual angle;
The training of multitask convolutional neural networks: multitask convolutional neural networks are multilayer convolution, and sample image is passed through original
Gray image and illumination invariant image after the conversion of beginning RGB image are more to train as the input of multitask convolutional neural networks
Task convolutional neural networks are trained image aspects information, vehicle number information and vehicle closeness information parallel;
Detection image: the multitask convolutional neural networks after image to be detected input training obtain and image to be detected
Corresponding image aspects information, vehicle number information and vehicle closeness information or combinations thereof information;
Vehicle queue length and congestion determine: comprehensive according to image aspects information, vehicle number information and vehicle closeness information
Conjunction judges vehicle congestion situation.
The direct method that vehicle congestion is detected using still image of the invention, utilizes multitask convolutional neural networks
Mode obtains multitask convolutional neural networks model by training, therefore in actually detected, is only captured by a road
Image can detect the situation of congestion in road.
Detailed description of the invention
Fig. 1 is the flow chart for the method that the present invention directly detects vehicle congestion using still image.
When Fig. 2 is multitask convolutional neural networks training in Fig. 1, the training method figure of multitask convolutional neural networks.
When Fig. 3 is data prediction in Fig. 1, the acquisition of image aspects information, vehicle number information and vehicle closeness information
Exemplary diagram.
When Fig. 4 is multitask convolutional neural networks training in Fig. 1, the input mode exemplary diagram of sample image.
When Fig. 5 is multitask convolutional neural networks training in Fig. 1, the exemplary diagram of the multitask convolutional neural networks of use.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.
The direct method that vehicle congestion is detected using still image of the invention, suitable for being executed in computer equipment,
It can be made the form of software installation packet in practical applications, also can be made the form of APP, such as cell phone application.
Referring to Fig. 1, the method for directly being detected vehicle congestion using still image is mainly comprised the steps that
Data prediction: width sample image every in training sample image data is divided into m × n subgraph, takes people
The form of work calibration is that each subgraph assigns its two attribute information: vehicle number information and vehicle closeness information;Original image
Image aspects information is assigned according to camera shooting grease head highness visual angle;
The training of multitask convolutional neural networks: multitask convolutional neural networks are multilayer convolution, and sample image is passed through original
Gray image and illumination invariant image after the conversion of beginning RGB image are more to train as the input of multitask convolutional neural networks
Task convolutional neural networks are trained image aspects information, vehicle number information and vehicle closeness information parallel;
Detection image: the multitask convolutional neural networks after image to be detected input training obtain and image to be detected
Corresponding image aspects information, vehicle number information and vehicle closeness information or combinations thereof information;
Vehicle queue length and congestion determine: comprehensive according to image aspects information, vehicle number information and vehicle closeness information
Conjunction judges vehicle congestion situation.
Incorporated by reference to Fig. 2, in the present case, multitask convolutional neural networks carry out convolution to both direction: a direction is figure
As Viewing-angle information, other direction is vehicle number information and vehicle closeness information.
The present invention obtains multitask convolutional neural networks mould in the way of multitask convolutional neural networks, through training
Type can only detect the situation of congestion in road in actually detected by the image that a road is captured.From current existing side
Case is seen, does not find also only to carry out the scheme that congestion in road detection determines by picture.
Subsequently, it gives one example and explanation is explained in detail.
1) data prediction
Width original image every in training sample image data is divided into 4 × 4 subgraphs, each subgraph tool there can be 2 kinds
Attribute information: vehicle number and vehicle closeness, wherein vehicle closeness can be determined by vehicle number.
Vehicle closeness can define 4 ranks: the corresponding vehicle closeness information of the vehicle number information of 0-5 vehicle is 0;6-
The corresponding vehicle closeness information of the vehicle number information of 10 vehicles is 1;The corresponding vehicle of vehicle number information of 11-15 vehicle is intensive
Spending information is 2;The corresponding vehicle closeness information of the vehicle number information of 16 vehicles is 3.
Image aspects information determines that image aspects information is according to road by the mounting height and shooting visual angle of road camera
The mounting height and shooting visual angle of camera are divided into basic, normal, high three kinds of visual angles.As shown in Figure 3, there are one figures for whole image
As Viewing-angle information, height and visual angle depending on road camera are divided into three kinds basic, normal, high.
2) multitask convolutional neural networks training
The training of multitask convolutional neural networks, which can be divided into, to be described in detail: multichannel input, multitask convolutional Neural net
Network.
Multichannel input: each training sample figure takes multichannel input mode, as shown in Figure 4.Multichannel is by original
The gray image and illumination invariant image composition of beginning RGB image conversion.Illumination invariant image can describe object well, be easy to
Detection.
Multitask convolutional neural networks: multitask convolutional neural networks framework is as shown in figure 5, vehicle count and image aspects
First share 3 convolutional layers, then to the two directions carry out convolution, the Training strategy of multitask convolutional neural networks and other
As the training of neural network, when loss function convergence, the model that training obtains is tested on verifying collection, for inspection
The result of sniffing accidentally is analyzed, and is focused to find out the image of some corresponding types in supplemental training according to the type of the image of mistake
It is added in training set, network is continued to train.In the present embodiment, it is defined respectively as loss function.
Vehicle technology loss function: vehicle count can regard depth regression problem as, and input is whole image, and output is not
With the vehicle number of subgraph.Therefore, it is intended that minimizing Euclidean loss
Wherein k is the index of subgraph, and Gk and Pk are the true of kth subgraph and prediction vehicle number.
Vehicle closeness loss function: vehicle closeness can be defined according to vehicle number, but can be in this network
Directly show that vehicle closeness is lost by neural network by defining loss function
Wherein, k is the index of subgraph, pkmFor the areal densities grade of kth subgraph.
The recurrence of vehicle number and traffic density horizontal classification share all deconvolution parameters, only different in interior lamination parameter.
Image aspects loss function: image aspects loss
Herein, C is the sum (being set to 3 as described above) of image aspects classification, yicAnd picRespectively be prediction and
True picture visual angle.
It is solved finally by majorized function, forms convolutional neural networks model.
Herein, N is the quantity of training image, and λ is for balancing different loss functions.W and b is model parameter.λ setting
For 0, the network will not consider image aspects.
3) detection image
Image to be detected is inputted, vehicle count process can be used only and obtain number of vehicles judgement vehicle congestion situation.
4) vehicle queue length and congestion determine
Vehicle congestion situation is judged according to image aspects information, vehicle number information and vehicle closeness informix.
Congestion determines that there are many methods, and number of vehicles threshold value such as is set separately according to unused view level, is more than threshold value
As congestion, or congestion can be determined with the ratio or index information of number of vehicles and Viewing-angle information.It is long as vehicle queue
Degree, in general, vehicle more multiple-length is longer, is not that say have to an exact length.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Made any modifications, equivalent replacements, and improvements etc., should all be included in the protection scope of the present invention within mind and principle.
Claims (8)
1. a kind of method that vehicle congestion is directly detected using still image, suitable for being executed in computer equipment, feature
Be: itself the following steps are included:
Data prediction: width sample image every in training sample image data is divided into m × n subgraph, takes artificial mark
Fixed form is that each subgraph assigns its two attribute information: vehicle number information and vehicle closeness information;Original image according to
It images grease head highness visual angle and assigns image aspects information;
The training of multitask convolutional neural networks: multitask convolutional neural networks are multilayer convolution, and sample image is passed through original RGB
Gray image and illumination invariant image after image conversion, as the input of multitask convolutional neural networks, Lai Xunlian multitask
Convolutional neural networks are trained image aspects information, vehicle number information and vehicle closeness information parallel;
Multitask convolutional neural networks carry out convolution to both direction: a direction is image aspects information, and other direction is vehicle
Number information and vehicle closeness information;
Loss function is defined in multitask convolutional neural networks:
One, vehicle count loss function: vehicle count is regarded as depth regression problem, input is whole image, and output is different
The vehicle number of subgraph minimizes Euclidean lossUsing formulaWherein, k is the index of subgraph, Gk
It is the true of kth subgraph and prediction vehicle number with Pk;
Two, vehicle closeness loss function: vehicle closeness is defined according to vehicle number, the loss of vehicle closenessUsing public affairs
FormulaWherein k is the index of subgraph, and pkm is the areal densities grade of kth subgraph;
Three, image aspects loss function: image aspects lossUsing
Wherein, C is the sum of image aspects classification, yicAnd pic
It is prediction and true picture visual angle respectively;
By above three loss function, the multitask convolutional neural networks after training obtain convolutional neural networks model:Wherein, N is the quantity of training sample image, and λ is for balancing different loss letters
Number, W and b are model parameters, and the multitask convolutional neural networks that λ is set as after 0 training will not consider image aspects;
Detection image: the multitask convolutional neural networks after image to be detected input training obtain opposite with image to be detected
Image aspects information, vehicle number information and the vehicle closeness information answered or combinations thereof information;
Vehicle queue length and congestion determine: being sentenced according to image aspects information, vehicle number information and vehicle closeness informix
Disconnected vehicle congestion situation.
2. the method as described in claim 1 for directly detecting vehicle congestion using still image, it is characterised in that: vehicle is close
Intensity information is determined by vehicle number information.
3. the method as claimed in claim 2 for directly detecting vehicle congestion using still image, it is characterised in that: vehicle is close
Intensity information defines 4 ranks: the corresponding vehicle closeness information of the vehicle number information of 0-5 vehicle is 0;The vehicle of 6-10 vehicle
The corresponding vehicle closeness information of number information is 1;The corresponding vehicle closeness information of the vehicle number information of 11-15 vehicle is 2;16
The corresponding vehicle closeness information of vehicle number information of vehicle is 3.
4. the method as described in claim 1 for directly detecting vehicle congestion using still image, it is characterised in that: image view
Angle information is determined by the mounting height and shooting visual angle of road camera.
5. the method as claimed in claim 4 for directly detecting vehicle congestion using still image, it is characterised in that: image view
Angle information is divided into basic, normal, high three kinds of visual angles according to the mounting height and shooting visual angle of road camera.
6. the method as described in claim 1 for directly detecting vehicle congestion using still image, it is characterised in that: multitask
Convolutional neural networks share 3 convolutional layers.
7. a kind of storage equipment, wherein storing a plurality of instruction, described instruction is suitable for processor and loads and execute, it is characterised in that:
The a plurality of instruction are as follows:
Data prediction: width sample image every in training sample image data is divided into m × n subgraph, takes artificial mark
Fixed form is that each subgraph assigns its two attribute information: vehicle number information and vehicle closeness information;Original image according to
It images grease head highness visual angle and assigns image aspects information;
The training of multitask convolutional neural networks: multitask convolutional neural networks are multilayer convolution, and sample image is passed through original RGB
Gray image and illumination invariant image after image conversion, as the input of multitask convolutional neural networks, Lai Xunlian multitask
Convolutional neural networks are trained image aspects information, vehicle number information and vehicle closeness information parallel;
Multitask convolutional neural networks carry out convolution to both direction: a direction is image aspects information, and other direction is vehicle
Number information and vehicle closeness information;
Loss function is defined in multitask convolutional neural networks:
One, vehicle count loss function: vehicle count is regarded as depth regression problem, input is whole image, and output is different
The vehicle number of subgraph minimizes Euclidean lossUsing formulaWherein, k is the index of subgraph, Gk
It is the true of kth subgraph and prediction vehicle number with Pk;
Two, vehicle closeness loss function: vehicle closeness is defined according to vehicle number, the loss of vehicle closenessUsing public affairs
FormulaWherein k is the index of subgraph, and pkm is the areal densities grade of kth subgraph;
Three, image aspects loss function: image aspects lossUsing
Wherein, C is the sum of image aspects classification, yicAnd pic
It is prediction and true picture visual angle respectively;
By above three loss function, the multitask convolutional neural networks after training obtain convolutional neural networks model:Wherein, N is the quantity of training sample image, and λ is for balancing different loss letters
Number, W and b are model parameters, and the multitask convolutional neural networks that λ is set as after 0 training will not consider image aspects;
Detection image: the multitask convolutional neural networks after image to be detected input training obtain opposite with image to be detected
Image aspects information, vehicle number information and the vehicle closeness information answered or combinations thereof information;
Vehicle queue length and congestion determine: being sentenced according to image aspects information, vehicle number information and vehicle closeness informix
Disconnected vehicle congestion situation.
8. a kind of mobile communication terminal comprising:
Processor is adapted for carrying out various instructions;And
Equipment is stored, is suitable for storing a plurality of instruction, described instruction is suitable for processor and loads and execute;
It is characterized by: a plurality of instruction are as follows:
Data prediction: width sample image every in training sample image data is divided into m × n subgraph, takes artificial mark
Fixed form is that each subgraph assigns its two attribute information: vehicle number information and vehicle closeness information;Original image according to
It images grease head highness visual angle and assigns image aspects information;
The training of multitask convolutional neural networks: multitask convolutional neural networks are multilayer convolution, and sample image is passed through original RGB
Gray image and illumination invariant image after image conversion, as the input of multitask convolutional neural networks, Lai Xunlian multitask
Convolutional neural networks are trained image aspects information, vehicle number information and vehicle closeness information parallel;
Multitask convolutional neural networks carry out convolution to both direction: a direction is image aspects information, and other direction is vehicle
Number information and vehicle closeness information;
Loss function is defined in multitask convolutional neural networks:
One, vehicle count loss function: vehicle count is regarded as depth regression problem, input is whole image, and output is different
The vehicle number of subgraph minimizes Euclidean lossUsing formulaWherein, k is the index of subgraph, Gk
It is the true of kth subgraph and prediction vehicle number with Pk;
Two, vehicle closeness loss function: vehicle closeness is defined according to vehicle number, the loss of vehicle closenessUsing public affairs
FormulaWherein k is the index of subgraph, and pkm is the areal densities grade of kth subgraph;
Three, image aspects loss function: image aspects lossUsing
Wherein, C is the sum of image aspects classification, yicAnd pic
It is prediction and true picture visual angle respectively;
By above three loss function, the multitask convolutional neural networks after training obtain convolutional neural networks model:Wherein, N is the quantity of training sample image, and λ is for balancing different loss letters
Number, W and b are model parameters, and the multitask convolutional neural networks that λ is set as after 0 training will not consider image aspects;
Detection image: the multitask convolutional neural networks after image to be detected input training obtain opposite with image to be detected
Image aspects information, vehicle number information and the vehicle closeness information answered or combinations thereof information;
Vehicle queue length and congestion determine: being sentenced according to image aspects information, vehicle number information and vehicle closeness informix
Disconnected vehicle congestion situation.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710339106.8A CN106971544B (en) | 2017-05-15 | 2017-05-15 | A kind of direct method that vehicle congestion is detected using still image |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710339106.8A CN106971544B (en) | 2017-05-15 | 2017-05-15 | A kind of direct method that vehicle congestion is detected using still image |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106971544A CN106971544A (en) | 2017-07-21 |
CN106971544B true CN106971544B (en) | 2019-07-16 |
Family
ID=59331648
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710339106.8A Active CN106971544B (en) | 2017-05-15 | 2017-05-15 | A kind of direct method that vehicle congestion is detected using still image |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106971544B (en) |
Families Citing this family (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107808140B (en) * | 2017-11-07 | 2020-07-31 | 浙江大学 | Monocular vision road recognition algorithm based on image fusion |
CN108039044B (en) * | 2017-12-05 | 2021-06-01 | 安徽大学 | Vehicle intelligent queuing system and method based on multi-scale convolutional neural network |
CN108550259B (en) * | 2018-04-19 | 2020-05-12 | 何澜 | Road congestion judging method, terminal device and computer readable storage medium |
CN109255374A (en) * | 2018-08-27 | 2019-01-22 | 中共中央办公厅电子科技学院 | A kind of aesthetic properties evaluation method based on intensive convolutional network and multitask network |
CN109858372B (en) * | 2018-12-29 | 2021-04-27 | 浙江零跑科技有限公司 | Lane-level precision automatic driving structured data analysis method |
CN109887276B (en) * | 2019-01-30 | 2020-11-03 | 北京同方软件有限公司 | Night traffic jam detection method based on fusion of foreground extraction and deep learning |
CN110390822A (en) * | 2019-05-31 | 2019-10-29 | 东南大学 | Bridge statistical method of traffic flow based on FBG sensor and convolutional neural networks |
CN112489456B (en) * | 2020-12-01 | 2022-01-28 | 山东交通学院 | Signal lamp regulation and control method and system based on urban trunk line vehicle queuing length |
CN113570858B (en) * | 2021-07-22 | 2022-10-21 | 吉林大学 | System and method for assisting vehicle to identify traffic jam condition by unmanned aerial vehicle |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104036323A (en) * | 2014-06-26 | 2014-09-10 | 叶茂 | Vehicle detection method based on convolutional neural network |
CN106023605A (en) * | 2016-07-15 | 2016-10-12 | 姹ゅ钩 | Traffic signal lamp control method based on deep convolution neural network |
CN106297297A (en) * | 2016-11-03 | 2017-01-04 | 成都通甲优博科技有限责任公司 | Traffic jam judging method based on degree of depth study |
CN106599869A (en) * | 2016-12-22 | 2017-04-26 | 安徽大学 | Vehicle attribute identification method based on multi-task convolutional neural network |
CN106611162A (en) * | 2016-12-20 | 2017-05-03 | 西安电子科技大学 | Method for real-time detection of road vehicle based on deep learning SSD frame |
CN106650913A (en) * | 2016-12-31 | 2017-05-10 | 中国科学技术大学 | Deep convolution neural network-based traffic flow density estimation method |
-
2017
- 2017-05-15 CN CN201710339106.8A patent/CN106971544B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104036323A (en) * | 2014-06-26 | 2014-09-10 | 叶茂 | Vehicle detection method based on convolutional neural network |
CN106023605A (en) * | 2016-07-15 | 2016-10-12 | 姹ゅ钩 | Traffic signal lamp control method based on deep convolution neural network |
CN106297297A (en) * | 2016-11-03 | 2017-01-04 | 成都通甲优博科技有限责任公司 | Traffic jam judging method based on degree of depth study |
CN106611162A (en) * | 2016-12-20 | 2017-05-03 | 西安电子科技大学 | Method for real-time detection of road vehicle based on deep learning SSD frame |
CN106599869A (en) * | 2016-12-22 | 2017-04-26 | 安徽大学 | Vehicle attribute identification method based on multi-task convolutional neural network |
CN106650913A (en) * | 2016-12-31 | 2017-05-10 | 中国科学技术大学 | Deep convolution neural network-based traffic flow density estimation method |
Non-Patent Citations (2)
Title |
---|
Single-Image Crowd Counting via Multi-Column Convolutional Neural Network;Yingying Zhang;《2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)》;20161212;第589-597页 |
城市交通运行状态分析与预测关键技术研究;刘张;《中国博士学位论文全文数据库工程科技Ⅱ辑》;20170215(第2期);C034-110页 |
Also Published As
Publication number | Publication date |
---|---|
CN106971544A (en) | 2017-07-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106971544B (en) | A kind of direct method that vehicle congestion is detected using still image | |
CN104751634B (en) | The integrated application method of freeway tunnel driving image acquisition information | |
Jain et al. | Road traffic congestion in the developing world | |
CN104303193B (en) | Target classification based on cluster | |
CN102867417B (en) | Taxi anti-forgery system and taxi anti-forgery method | |
CN106530730A (en) | Traffic violation detection method and system | |
CN109615864A (en) | Vehicle congestion analysis method, system, terminal and storage medium based on video structural | |
KR20180046798A (en) | Method and apparatus for real time traffic information provision | |
CN103208185A (en) | Method and system for nighttime vehicle detection on basis of vehicle light identification | |
CN110660222A (en) | Intelligent environment-friendly electronic snapshot system for black smoke vehicle on road | |
CN105354530A (en) | Vehicle body color identification method and apparatus | |
CN106846813A (en) | The method for building urban road vehicle image data base | |
CN103729863B (en) | Traffic lights automatic location based on autonomous learning is known method for distinguishing | |
CN110415555A (en) | A kind of recognition methods of effective scribing line parking stall and system based on deep learning | |
CN109784254A (en) | A kind of method, apparatus and electronic equipment of rule-breaking vehicle event detection | |
WO2018157056A1 (en) | Video data background tracking and subtraction with multiple layers of stationary foreground and regions | |
CN107689158A (en) | A kind of intellectual traffic control method based on image procossing | |
CN112651293B (en) | Video detection method for road illegal spreading event | |
CN106570147A (en) | GIS road network analysis-based jump type video tracking method and system | |
CN110490150A (en) | A kind of automatic auditing system of picture violating the regulations and method based on vehicle retrieval | |
CN116153086B (en) | Multi-path traffic accident and congestion detection method and system based on deep learning | |
CN109903574A (en) | The acquisition methods and device of crossing traffic information | |
CN110163109A (en) | A kind of lane line mask method and device | |
Magrini et al. | Computer vision on embedded sensors for traffic flow monitoring | |
Zhou et al. | Method for judging parking status based on yolov2 target detection algorithm |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |