CN110458370A - A kind of urban roads safety prediction technique and system based on open data - Google Patents
A kind of urban roads safety prediction technique and system based on open data Download PDFInfo
- Publication number
- CN110458370A CN110458370A CN201910762822.6A CN201910762822A CN110458370A CN 110458370 A CN110458370 A CN 110458370A CN 201910762822 A CN201910762822 A CN 201910762822A CN 110458370 A CN110458370 A CN 110458370A
- Authority
- CN
- China
- Prior art keywords
- image
- traffic accident
- satellite image
- urban roads
- network model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
- G06Q50/265—Personal security, identity or safety
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- General Physics & Mathematics (AREA)
- Tourism & Hospitality (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- General Health & Medical Sciences (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Health & Medical Sciences (AREA)
- Development Economics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Computer Security & Cryptography (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Primary Health Care (AREA)
- Educational Administration (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention discloses a kind of urban roads safety prediction techniques and system based on open data, are related to technical field of computer vision.Satellite image and traffic accident point information are collected by open data platform first, and accident point is labeled on the corresponding position of satellite image according to traffic accident point information, is constituted and satellite image traffic accident point image of the same size；Then neural network model is generated using satellite image and traffic accident point image one sparse depth of training, until network model is restrained；Finally neural network model is generated using trained sparse depth to predict the road safety in other cities.The present invention uses open data, overcome the previous data for predicting that road safety faces problem at high cost, and it only needs for the satellite image for surveying city to be input to trained sparse depth and generates in neural network model, it can predict the position that the urban traffic accident occurs, to substantially increase forecasting efficiency.
Description
Technical field
The present invention relates to technical field of computer vision, more particularly to a kind of urban roads safety based on open data
Prediction technique and system.
Background technique
With the rapid development of modern city, Traffic Systems become particularly important, and traffic accident is obstruction system
The key factor of normal operation.Traffic accident also brings huge loss.The prevention of traffic accident becomes public safety
The important topic in field.In order to prevent traffic accident, many scholars are studied.But due to causing each side of traffic accident
The shortage of face data causes the research for preventing traffic accident to be difficult to promote.Recently as big data technology, technology of Internet of things and
The development of GIS-Geographic Information System so that the generation effectively to be avoided traffic accident using the mining analysis of traffic accident big data at
It is possible.
In traditional some researchs, traffic accident is prevented by crossing monitoring mechanism.Firstly, to the fixed view in a crossing
The monitoring video at angle is handled, and obtains vehicle characteristics and track in picture；Then, it can be predicted using feature and track training one
The neural network of vehicle traveling mode；Finally, feature and track acquisition are carried out to the different vehicle for entering the same crossing, by him
Prediction mode be input in a conditional probability model, predict a possibility that two vehicles collide.Such methods
It can only monitor that one crossing of prediction costs dearly and cumbersome if it is being widely used in city scope.
In other researchs, the corresponding relationship between the factor of traffic accident and accident is caused to predict traffic by discussion
Accident.Firstly, collecting a large amount of traffic accident information, including time, temperature, wind speed, condition of road surface, road speed limit etc.；Then,
These features relevant to traffic accident are inputted into certain classifier, such as support vector machines, decision tree, random forest or nerve
Network, training classifier is until convergence；Finally, the correlative factor in some section or place is input to classifier, this is predicted
Whether place occurs traffic accident.Such methods can widely predict the traffic accident possibility in section or place, but phase
The collection of pass factor is often highly difficult.
Summary of the invention
In order to solve the technical issues of above-mentioned background technique is mentioned, the present invention provides a kind of cities based on open data
Road safety prediction technique and system make the forecast cost of traffic accident lower, more efficient.
To achieve the above object, the present invention provides following schemes:
A kind of urban roads safety prediction technique based on open data, comprising:
It obtains the open data of urban road and is pre-processed；The urban road open data packet includes satellite image and friendship
Logical accident point information；
Using Density Estimator algorithm, by the traffic accident point information in conjunction with pretreated satellite image, generate
Traffic accident point image；
Image generates in neural network model with the training sparse depth sparse depth that is sequentially inputted to construct
Degree generates neural network model, obtains urban roads safety prediction model；The input of the urban roads safety prediction model is
Pretreated satellite image, the output of the urban roads safety prediction model are traffic accident point image；Described image pair
Including pretreated satellite image and the corresponding traffic accident point image of pretreated satellite image；
It obtains the satellite image in actual measurement city and is pre-processed；
The satellite image in pretreated actual measurement city is input in the urban roads safety prediction model to obtain
The traffic accident point image in city is surveyed, and traffic thing is occurred according to the traffic accident point image prediction actual measurement city in actual measurement city
Therefore location information.
Optionally, the acquisition urban road opens data and is pre-processed, and specifically includes:
Obtain the open data of urban road；The urban road open data packet includes satellite image and traffic accident point letter
Breath；
Scaling processing is merged to the satellite image, obtains pretreated satellite image.
Optionally, described to use Density Estimator algorithm, by the traffic accident point information and pretreated satellite mapping
As combining, generation traffic accident point image is specifically included:
Extract the GPS coordinate range in pretreated satellite image region；
According to the traffic accident point information, the location information that traffic accident occurs within the scope of the GPS coordinate is determined；
Using Density Estimator algorithm, the location information that traffic accident occurs is labeled in pretreated satellite image
On, generate the corresponding traffic accident point image of pretreated satellite image.
Optionally, described to generate image in neural network model the sparse depth for being sequentially inputted to construct with training
The sparse depth generates neural network model, obtains urban roads safety prediction model, specifically includes:
Building multiple images to and by described image to be divided into training image to and test image pair；
The training image generates in neural network model to adjust the sparse depth for being sequentially inputted to construct
The weight that sparse depth generates neural network model is stated, preliminary urban roads safety prediction model is obtained；The sparse depth is raw
Basic network model at neural network model is pix2pixHD network model, and the sparse depth generates neural network model
Loss function be sparse constraint loss function；
By the test image being sequentially inputted in preliminary urban roads safety prediction model with the determination preliminary city
The accuracy rate of city's road safety prediction model, and when the accuracy rate of the preliminary urban roads safety prediction model is less than setting standard
Sparse depth described in re -training generates neural network model when true rate, when the standard of the preliminary urban roads safety prediction model
When true rate is more than or equal to the setting accuracy rate, the preliminary urban roads safety prediction model is determined as final city
City's road safety prediction model.
Optionally, described to obtain the satellite image for surveying city and pre-processed, it specifically includes:
The satellite image in actual measurement city is obtained, and scaling processing is merged to the satellite image in actual measurement city, is obtained pre-
Treated actual measurement city satellite image.
A kind of urban roads safety forecasting system based on open data, comprising:
First obtains module, for obtaining the open data of urban road and being pre-processed；The open number of the urban road
According to including satellite image and traffic accident point information；
Traffic accident point image generation module, for use Density Estimator algorithm, by the traffic accident point information with
Pretreated satellite image combines, and generates traffic accident point image；
Urban roads safety prediction model determining module, for image is raw to the sparse depth for being sequentially inputted to construct
Neural network model is generated with the training sparse depth in neural network model, obtains urban roads safety prediction model；
The input of the urban roads safety prediction model is pretreated satellite image, the urban roads safety prediction model
Output is traffic accident point image；Described image is to including pretreated satellite image and pretreated satellite image pair
The traffic accident point image answered；
Second obtains module, for obtaining the satellite image in actual measurement city and being pre-processed；
Traffic accident location information prediction module, it is described for the satellite image in pretreated actual measurement city to be input to
The traffic accident point image in actual measurement city is obtained in urban roads safety prediction model, and according to the traffic accident in actual measurement city
The location information of traffic accident occurs for point image prediction actual measurement city.
Optionally, described first module is obtained, specifically included:
Acquiring unit, for obtaining the open data of urban road；The urban road open data packet include satellite image and
Traffic accident point information；
First pretreatment unit obtains pretreated satellite for merging scaling processing to the satellite image
Image.
Optionally, the traffic accident point image generation module, specifically includes:
Extraction unit, for extracting the GPS coordinate range in pretreated satellite image region；
Location information determination unit occurs for according to the traffic accident point information, determining within the scope of the GPS coordinate
The location information of traffic accident；
The location information that traffic accident occurs is labeled in pretreatment for using Density Estimator algorithm by generation unit
On satellite image afterwards, the corresponding traffic accident point image of pretreated satellite image is generated.
Optionally, the urban roads safety prediction model determining module, specifically includes:
Image to building division unit, for construct multiple images to and by described image to be divided into training image to
Test image pair；
Training unit, for the training image to be generated neural network mould to the sparse depth for being sequentially inputted to construct
To adjust the weight that the sparse depth generates neural network model in type, preliminary urban roads safety prediction model is obtained；Institute
Stating sparse depth and generating the basic network model of neural network model is pix2pixHD network model, and the sparse depth generates
The loss function of neural network model is sparse constraint loss function；
Testing and debugging unit, for by the test image to being sequentially inputted in preliminary urban roads safety prediction model
With the accuracy rate of the determination preliminary urban roads safety prediction model, and when the preliminary urban roads safety prediction model
Sparse depth described in re -training generates neural network model when accuracy rate is less than setting accuracy rate, when the preliminary urban road
When the accuracy rate of safe prediction model is more than or equal to the setting accuracy rate, the preliminary urban roads safety is predicted into mould
Type is determined as final urban roads safety prediction model.
Optionally, described second module is obtained, specifically included:
Second pretreatment unit is carried out for obtaining the satellite image in actual measurement city, and to the satellite image in actual measurement city
Merge scaling processing, obtains the satellite image in pretreated actual measurement city.
The specific embodiment provided according to the present invention, the invention discloses following technical effects:
The present invention provides a kind of urban roads safety prediction techniques and system based on open data.Pass through opening first
Data platform collects satellite image and traffic accident point information, and accident point is labeled in satellite mapping according to traffic accident point information
As constituting and satellite image traffic accident point image of the same size on corresponding position；Then satellite image and traffic are used
Accident point image one sparse depth of training generates neural network model, until network model is restrained；Finally using trained
Sparse depth generates neural network model to predict the road safety in other cities.The present invention is overcome in the past using open data
The data that prediction road safety faces problem at high cost, and the present invention only needs the satellite image for surveying city being input to instruction
The sparse depth perfected generates the position that can predict that the urban traffic accident occurs in neural network model, substantially increases pre-
Survey efficiency；In addition, have High relevancy between satellite image and traffic accident, it can be with table by satellite image visual signature abundant
A possibility that traffic accident occurs is levied, ensure that the accuracy of prediction.
Detailed description of the invention
It in order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, below will be to institute in embodiment
Attached drawing to be used is needed to be briefly described, it should be apparent that, the accompanying drawings in the following description is only some implementations of the invention
Example, for those of ordinary skill in the art, without any creative labor, can also be according to these attached drawings
Obtain other attached drawings.
Fig. 1 is the flow diagram of urban roads safety prediction technique of the embodiment of the present invention based on open data；
Fig. 2 is the whole legend flow chart of urban roads safety prediction technique of the embodiment of the present invention based on open data；
Fig. 3 is that the embodiment of the present invention inputs the image of sparse depth generation neural network model to schematic diagram；Wherein, Fig. 3
(a) indicate that pretreated satellite image, Fig. 3 (b) indicate traffic accident point image；
Fig. 4 is the test result figure in the city B of the embodiment of the present invention；Fig. 4 (a) indicates the first test result figure, Fig. 4 (b) table
Show that the second test result figure, Fig. 4 (c) indicate that third test result figure, Fig. 4 (d) indicate the 4th test result figure；
Fig. 5 is the structural schematic diagram of urban roads safety forecasting system of the embodiment of the present invention based on open data.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
The object of the present invention is to provide a kind of urban roads safety prediction technique and system based on open data, traffic is allowed
The forecast cost of accident is lower, more efficient.
In order to make the foregoing objectives, features and advantages of the present invention clearer and more comprehensible, with reference to the accompanying drawing and specific real
Applying mode, the present invention is described in further detail.
Embodiment one
As depicted in figs. 1 and 2, a kind of urban roads safety prediction technique based on open data provided in this embodiment,
Include the following steps.
Step 101: obtaining the open data of urban road and pre-processed；The urban road open data packet includes satellite
Image and traffic accident point information (also known as traffic accident location information).
Step 102: Density Estimator algorithm is used, by the traffic accident point information and pretreated satellite image knot
It closes, generates traffic accident point image.
Step 103: image generates in neural network model to train the sparse depth for being sequentially inputted to construct
It states sparse depth and generates neural network model, obtain urban roads safety prediction model；The urban roads safety prediction model
Input be pretreated satellite image, the output of the urban roads safety prediction model is traffic accident point image；Institute
Image is stated to including pretreated satellite image and the corresponding traffic accident point image of pretreated satellite image.
Step 104: obtaining the satellite image in actual measurement city and pre-processed.
Step 105: the satellite image in pretreated actual measurement city is input to the urban roads safety prediction model
In with obtain actual measurement city traffic accident point image, and according to actual measurement city traffic accident point image prediction actual measurement city hair
The location information of raw traffic accident.
Further, in a step 101, the original number for using the open data of urban road to predict as urban roads safety
According to.
In this embodiment, the open data of road of two different cities are had collected.Road open data packet contains satellite
Image and traffic accident point information, the rich visual features that satellite image has can characterize a possibility that traffic accident occurs,
And the traffic accident point information downloaded to, it needs to be labeled on the corresponding place of satellite image.Wherein, the satellite mapping downloaded
It seem the satellite image tile of the 256*256 of Level-19, the open data of the road in the city A are used to train, and total 37K；The city B
The open data of road be used to test.
Since the sparse depth generation neural network model constructed is bigger, and GPU memory is fixed, and input figure is limited
The size of picture.In order to input bigger image as far as possible, to obtain more global informations, without losing characteristics of image, use
Satellite image tile 5*5 is merged into big image and then zooms to the scheme of the size of 800*800.So carrying out network mould
Before type training test, scaling processing first is merged to satellite image to obtain pretreated satellite image, is realized to city
The pretreatment of the open data of city's road.
Further, in a step 102, using Density Estimator algorithm by traffic accident point information labeling after the pre-treatment
On satellite image, generate corresponding traffic accident point image, form semantic tagger, finally by pretreated satellite image and
Corresponding traffic accident point image generates the input of neural network model as next step sparse depth.
In order to make the traffic accident point of mark more meet the actual distribution of accident, traffic is generated using Density Estimator algorithm
Accident point image is as follows with the specific processing method for constituting semantic tagger:
(1) the GPS coordinate range in pretreated satellite image region is extracted.
(2) according to traffic accident point information, the location information that traffic accident occurs within the scope of the GPS coordinate is determined.
(3) Density Estimator algorithm is used, the location information that traffic accident occurs is labeled in pretreated satellite mapping
As upper, corresponding traffic accident point image is generated.
Density Estimator calculating is carried out using the scientific algorithm library scipy of python in this example, detailed process is as follows:
(3.1) above-mentioned data are handled using Density Estimator function, obtains Density Estimator result；Density Estimator function
It is scipy.stats.gaussian_kde (), specifically how uses, is discussed in detail in the source code of the function, herein no longer
It discusses；In addition, the Density Estimator obtained is the result is that numerical matrix herein.
(3.2) by Density Estimator result by the matplotlib.image image library of python, color combining image,
It is converted into gray level image.
(3.3) it is scaled finally by gray level image, obtains the traffic thing with pretreated satellite image equal resolution
Therefore point image.
Obtained traffic accident point image is handled as shown in figure 3, left side a image is pretreated satellite image, right side b
Image is the corresponding traffic accident point image in the region, and the size of stain represents the region that traffic accident may occur in image
A possibility that size, shade represents traffic accident, is just.
Further, in step 103, first building multiple images to and by image to be divided into training image to and test
Image pair；Secondly, setting learning rate and the number of iterations, training image generates the sparse depth for being sequentially inputted to construct
To adjust the weight that sparse depth generates neural network model in neural network model, training obtains preliminary until network convergence
Urban roads safety prediction model；Sparse depth generates neural network model using pix2pixHD network model as basic network
Model, the loss of addition sparse constraint is built-up on loss function, i.e. the facilities network of sparse depth generation neural network model
Network model is pix2pixHD network model, and loss function is sparse constraint loss function；Finally by test image to successively defeated
Enter into preliminary urban roads safety prediction model with the accuracy rate of the preliminary urban roads safety prediction model of determination, and when preliminary
Re -training sparse depth generates neural network model when the accuracy rate of urban roads safety prediction model is less than setting accuracy rate,
When the accuracy rate of preliminary urban roads safety prediction model is more than or equal to setting accuracy rate, by preliminary urban roads safety
Prediction model is determined as final urban roads safety prediction model.
The present invention uses pix2pixHD network model, improves loss function to adapt to generate target.Original loss letter
Number is:
Wherein, in braces first item be generator loss, Section 2 is the loss of arbiter characteristic matching, is totally contained
Justice is that generator and the loss of arbiter characteristic matching are minimized while maximizing arbiter loss.Due to target of the present invention
Sparsity joined sparse constraint loss to improve network model performance.Sparse constraint loss is expressed as follows: LS(G)=| | G |
|1。
G is the image that generator generates, LSIt is 1 norm of image.Because of accident point institute's accounting in entire satellite image
Example very little, sparse constraint improve the integrality for generating accident point in image, reduce false-alarm.Finally, improved loss letter
Number is as follows:
When training, downloading sparse depth generates neural network model, sets learning rate and the number of iterations, input picture
It is right, it is then finely adjusted on its basis, until convergence.Trained sparse depth generates neural network model at this time
Predict the road safety in other cities.
Further, at step 104, the open data of the road for surveying city are also done into corresponding pretreatment, i.e., to actual measurement
The satellite image in city merges scaling processing, obtains the satellite image in pretreated actual measurement city.
Further, in step 105, the satellite image in pretreated actual measurement city is input to trained sparse depth
Degree generates neural network model, and prediction obtains the corresponding traffic accident point image of the satellite image, and then predicts actual measurement city
The location information of traffic accident occurs.
Embodiment two
The present embodiment also provides a kind of urban roads safety prediction technique based on open data, comprising: the training stage,
Test phase and actual measurement stage.
Training stage, including
(1) the Level-19 grade satellite image in the city A, total 37K open data downloading: have been downloaded by ground image website
, it is the color image of RGB triple channel that resolution ratio, which is 256*256, and the upper left corner of every image and the lower right corner are subsidiary
GPS coordinate；2012~2018 year traffic accident statistics table of the city A is downloaded to by the open data platform in the city A, includes traffic thing
Therefore the time occurred and position (GPS coordinate).
(2) data prediction: the 37K downloaded to satellite images may be constructed the satellite image tile square of 170*220
Battle array, in order to obtain more global informations without losing characteristics of image, by satellite image tile matrix with every 5*5 satellite images
Combined mode merges, and constitutes block image, block image is then zoomed to 800*800 image in different resolution to adapt to
GPU memory.It finally obtained 34*44 training images, resolution ratio is 800*800, is RGB Three Channel Color image.Entirely
Process may be formulated are as follows:
Wherein, w indicates that original satellite image tiles matrix data collection, Γ indicate the training image data after data prediction
Collection, the upper right footmark of each data set indicate the size (long * wide) of data set, and bottom right footmark indicates each image in data set
Size (channel long * wide *).While calculation block image, the traffic story point within the scope of the block is counted, is then used
Density Estimator algorithm generates the corresponding traffic accident point image of the block, may finally obtain 34*44 to training image pair.
(3) training pattern: using pix2pixHD network model to generate neural network model as depth in this example, comes
Feature connection between the satellite image and traffic accident point image of learning training image pair.Pix2pixHD network model is
A kind of condition generation confrontation network model for realizing image interpretation, it passes through the continuous game of generator G and arbiter D, to mention
The performance of high whole network model.Specifically for this example, the satellite image of generator input training image centering, then
Characteristics of image is obtained by convolution, then generates traffic accident point image by deconvolution；And arbiter inputs original image pair
(corresponding training image to) and image pair is generated, it is whether true to nature to judge to generate image.In training, setting network model
Input and output size is 800*800, and learning rate 0.00001, the constant the number of iterations of learning rate is 50 times, learning rate decaying iteration
Number is 50 times.After training, which can automatically save the weight of generator and arbiter, and in test phase and actual measurement
Stage only needs the weight of generator.
Test phase, comprising:
(1) open data downloading: by method identical with the training stage, the satellite image in the city B is downloaded.
(2) data prediction: by method identical with the training stage, the satellite image in the city B is handled.
(3) it is loaded into model: being predicted using pix2pixHD network model, need to be loaded into trained weight.Although
Training stage has trained generator and arbiter simultaneously, but only needs to carry out image interpretation using generator in test phase.
Therefore it may only be necessary to which the weight of generator is loaded into pix2pixHD network model, then by treated, satellite image passes through generation
Device, so that it may the traffic accident point image predicted.
(4) it predicts: the satellite image pre-processed being inputted to the generator for being loaded into weight, so that it may the traffic predicted
Accident point image.Test results are shown in figure 4 on the city B, and left, center, right respectively indicates satellite image, mark collision diagram
Picture and prediction traffic-accident image.
Test phase is for judging whether the weight of generator is suitable.
The actual measurement stage, comprising:
(1) acquisition is handled: being obtained the satellite image in actual measurement city and is pre-processed.
(2) it predicts: the satellite image in pretreated actual measurement city is input in trained model to obtain actual measurement
The traffic accident point image in city, and traffic accident is occurred according to the traffic accident point image prediction actual measurement city in actual measurement city
Location information.
Embodiment three
As shown in figure 5, present embodiments providing a kind of urban roads safety forecasting system based on open data, comprising:
First obtains module 501, for obtaining the open data of urban road and being pre-processed；The urban road is open
Data include satellite image and traffic accident point information.
Traffic accident point image generation module 502, for using Density Estimator algorithm, by the traffic accident point information
In conjunction with pretreated satellite image, traffic accident point image is generated.
Urban roads safety prediction model determining module 503, for by image to the sparse depth for being sequentially inputted to construct
Degree generates in neural network model and generates neural network model with the training sparse depth, obtains urban roads safety prediction mould
Type；The input of the urban roads safety prediction model is pretreated satellite image, and the urban roads safety predicts mould
The output of type is traffic accident point image；Described image is to including pretreated satellite image and pretreated satellite mapping
As corresponding traffic accident point image.
Second obtains module 504, for obtaining the satellite image in actual measurement city and being pre-processed.
Traffic accident location information prediction module 505, for the satellite image in pretreated actual measurement city to be input to
The traffic accident point image in actual measurement city is obtained in the urban roads safety prediction model, and according to the traffic in actual measurement city
Accident point image prediction surveys the location information that traffic accident occurs for city.
Described first obtains module 501, specifically includes:
Acquiring unit, for obtaining the open data of urban road；The urban road open data packet include satellite image and
Traffic accident point information.
First pretreatment unit obtains pretreated satellite for merging scaling processing to the satellite image
Image.
The traffic accident point image generation module 502, specifically includes:
Extraction unit, for extracting the GPS coordinate range in pretreated satellite image region.
Location information determination unit occurs for according to the traffic accident point information, determining within the scope of the GPS coordinate
The location information of traffic accident.
The location information that traffic accident occurs is labeled in pretreatment for using Density Estimator algorithm by generation unit
On satellite image afterwards, the corresponding traffic accident point image of pretreated satellite image is generated.
The urban roads safety prediction model determining module 503, specifically includes:
Image to building division unit, for construct multiple images to and by described image to be divided into training image to
Test image pair.
Training unit, for the training image to be generated neural network mould to the sparse depth for being sequentially inputted to construct
To adjust the weight that the sparse depth generates neural network model in type, preliminary urban roads safety prediction model is obtained；Institute
Stating sparse depth and generating the basic network model of neural network model is pix2pixHD network model, and the sparse depth generates
The loss function of neural network model is sparse constraint loss function.
Testing and debugging unit, for by the test image to being sequentially inputted in preliminary urban roads safety prediction model
With the accuracy rate of the determination preliminary urban roads safety prediction model, and when the preliminary urban roads safety prediction model
Sparse depth described in re -training generates neural network model when accuracy rate is less than setting accuracy rate, when the preliminary urban road
When the accuracy rate of safe prediction model is more than or equal to the setting accuracy rate, the preliminary urban roads safety is predicted into mould
Type is determined as final urban roads safety prediction model.
Described second obtains module 504, specifically includes:
Second pretreatment unit is carried out for obtaining the satellite image in actual measurement city, and to the satellite image in actual measurement city
Merge scaling processing, obtains the satellite image in pretreated actual measurement city.
Each embodiment in this specification is described in a progressive manner, the highlights of each of the examples are with other
The difference of embodiment, the same or similar parts in each embodiment may refer to each other.For system disclosed in embodiment
For, since it is corresponded to the methods disclosed in the examples, so being described relatively simple, related place is said referring to method part
It is bright.
Used herein a specific example illustrates the principle and implementation of the invention, and above embodiments are said
It is bright to be merely used to help understand method and its core concept of the invention；At the same time, for those skilled in the art, foundation
Thought of the invention, there will be changes in the specific implementation manner and application range.In conclusion the content of the present specification is not
It is interpreted as limitation of the present invention.
Claims (10)
1. a kind of urban roads safety prediction technique based on open data, which is characterized in that the urban roads safety prediction
Method, comprising:
It obtains the open data of urban road and is pre-processed；The urban road open data packet includes satellite image and traffic thing
Therefore point information；
Using Density Estimator algorithm, by the traffic accident point information in conjunction with pretreated satellite image, traffic is generated
Accident point image；
Image generates the sparse depth that is sequentially inputted to construct raw with the training sparse depth in neural network model
At neural network model, urban roads safety prediction model is obtained；The input of the urban roads safety prediction model is pre- place
Satellite image after reason, the output of the urban roads safety prediction model are traffic accident point image；Described image is to including
Pretreated satellite image and the corresponding traffic accident point image of pretreated satellite image；
It obtains the satellite image in actual measurement city and is pre-processed；
The satellite image in pretreated actual measurement city is input in the urban roads safety prediction model to obtain actual measurement
The traffic accident point image in city, and traffic accident is occurred according to the traffic accident point image prediction actual measurement city in actual measurement city
Location information.
2. the urban roads safety prediction technique according to claim 1 based on open data, which is characterized in that described to obtain
It takes the open data of urban road and is pre-processed, specifically included:
Obtain the open data of urban road；The urban road open data packet includes satellite image and traffic accident point information；
Scaling processing is merged to the satellite image, obtains pretreated satellite image.
3. the urban roads safety prediction technique according to claim 1 based on open data, which is characterized in that described to adopt
Traffic accident point is generated by the traffic accident point information in conjunction with pretreated satellite image with Density Estimator algorithm
Image specifically includes:
Extract the GPS coordinate range in pretreated satellite image region；
According to the traffic accident point information, the location information that traffic accident occurs within the scope of the GPS coordinate is determined；
Using Density Estimator algorithm, the location information that traffic accident occurs is labeled on pretreated satellite image, it is raw
At the corresponding traffic accident point image of pretreated satellite image.
4. the urban roads safety prediction technique according to claim 1 based on opening data, which is characterized in that described to incite somebody to action
Image generates in neural network model the sparse depth that is sequentially inputted to construct and generates nerve with the training sparse depth
Network model obtains urban roads safety prediction model, specifically includes:
Building multiple images to and by described image to be divided into training image to and test image pair；
The training image generates the sparse depth for being sequentially inputted to construct described dilute to adjust in neural network model
Degree of deepening by dredging generates the weight of neural network model, obtains preliminary urban roads safety prediction model；The sparse depth generates mind
Basic network model through network model is pix2pixHD network model, and the sparse depth generates the damage of neural network model
Mistake function is sparse constraint loss function；
By the test image being sequentially inputted in preliminary urban roads safety prediction model with the determination preliminary city road
The accuracy rate of road safe prediction model, and when the accuracy rate of the preliminary urban roads safety prediction model is less than setting accuracy rate
When re -training described in sparse depth generate neural network model, when the accuracy rate of the preliminary urban roads safety prediction model
When more than or equal to the setting accuracy rate, the preliminary urban roads safety prediction model is determined as final city road
Road safe prediction model.
5. the urban roads safety prediction technique according to claim 1 based on open data, which is characterized in that described to obtain
It takes the satellite image in actual measurement city and is pre-processed, specifically included:
The satellite image in actual measurement city is obtained, and scaling processing is merged to the satellite image in actual measurement city, is pre-processed
The satellite image in actual measurement city afterwards.
6. a kind of urban roads safety forecasting system based on open data, which is characterized in that the urban roads safety prediction
System, comprising:
First obtains module, for obtaining the open data of urban road and being pre-processed；The urban road open data packet
Include satellite image and traffic accident point information；
Traffic accident point image generation module, for using Density Estimator algorithm, by the traffic accident point information and pre- place
Satellite image after reason combines, and generates traffic accident point image；
Urban roads safety prediction model determining module, for image to be generated mind to the sparse depth for being sequentially inputted to construct
Through generating neural network model in network model with the training sparse depth, urban roads safety prediction model is obtained；It is described
The input of urban roads safety prediction model is pretreated satellite image, the output of the urban roads safety prediction model
For traffic accident point image；Described image is to corresponding including pretreated satellite image and pretreated satellite image
Traffic accident point image；
Second obtains module, for obtaining the satellite image in actual measurement city and being pre-processed；
Traffic accident location information prediction module, for the satellite image in pretreated actual measurement city to be input to the city
The traffic accident point image in actual measurement city is obtained in road safety prediction model, and according to the traffic accident point diagram in actual measurement city
As the location information of traffic accident occurs for prediction actual measurement city.
7. the urban roads safety forecasting system according to claim 6 based on open data, which is characterized in that described the
One obtains module, specifically includes:
Acquiring unit, for obtaining the open data of urban road；The urban road open data packet includes satellite image and traffic
Accident point information；
First pretreatment unit obtains pretreated satellite image for merging scaling processing to the satellite image.
8. the urban roads safety forecasting system according to claim 6 based on open data, which is characterized in that the friendship
Interpreter's event point image generation module, specifically includes:
Extraction unit, for extracting the GPS coordinate range in pretreated satellite image region；
Location information determination unit, for according to the traffic accident point information, determining generation traffic within the scope of the GPS coordinate
The location information of accident；
The location information that traffic accident occurs is labeled in pretreated by generation unit for using Density Estimator algorithm
On satellite image, the corresponding traffic accident point image of pretreated satellite image is generated.
9. the urban roads safety forecasting system according to claim 6 based on open data, which is characterized in that the city
City's road safety prediction model determining module, specifically includes:
Image to building division unit, for construct multiple images to and by described image to be divided into training image to and test
Image pair；
Training unit, for generating in neural network model the training image to the sparse depth for being sequentially inputted to construct
To adjust the weight that the sparse depth generates neural network model, preliminary urban roads safety prediction model is obtained；It is described dilute
The basic network model that degree of deepening by dredging generates neural network model is pix2pixHD network model, and the sparse depth generates nerve
The loss function of network model is sparse constraint loss function；
Testing and debugging unit, for by the test image to being sequentially inputted in preliminary urban roads safety prediction model with true
The accuracy rate of the fixed preliminary urban roads safety prediction model, and work as the accurate of the preliminary urban roads safety prediction model
Sparse depth described in re -training generates neural network model when rate is less than setting accuracy rate, when the preliminary urban roads safety
It is when the accuracy rate of prediction model is more than or equal to the setting accuracy rate, the preliminary urban roads safety prediction model is true
It is set to final urban roads safety prediction model.
10. the urban roads safety forecasting system according to claim 6 based on open data, which is characterized in that described
Second obtains module, specifically includes:
Second pretreatment unit is merged for obtaining the satellite image in actual measurement city, and to the satellite image in actual measurement city
Scaling processing obtains the satellite image in pretreated actual measurement city.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910762822.6A CN110458370A (en) | 2019-08-19 | 2019-08-19 | A kind of urban roads safety prediction technique and system based on open data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910762822.6A CN110458370A (en) | 2019-08-19 | 2019-08-19 | A kind of urban roads safety prediction technique and system based on open data |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110458370A true CN110458370A (en) | 2019-11-15 |
Family
ID=68487552
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910762822.6A Pending CN110458370A (en) | 2019-08-19 | 2019-08-19 | A kind of urban roads safety prediction technique and system based on open data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110458370A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111028512A (en) * | 2019-12-31 | 2020-04-17 | 福建工程学院 | Real-time traffic prediction method and device based on sparse BP neural network |
| CN111723997A (en) * | 2020-06-23 | 2020-09-29 | 上海电科智能系统股份有限公司 | Automatic generation method of urban major traffic accident data sample based on GAN |
| CN115018318A (en) * | 2022-06-01 | 2022-09-06 | 航天神舟智慧系统技术有限公司 | Method and system for predicting and analyzing risk of social area |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101976501A (en) * | 2010-10-29 | 2011-02-16 | 天津市市政工程设计研究院 | Principal component analysis and neural network based port road safety prediction method |
| CN104599443A (en) * | 2015-01-12 | 2015-05-06 | 江苏省交通规划设计院股份有限公司 | Vehicle-mounted forewarning terminal for driving behaviors based on information fusion and forewarning method thereof |
| CN110009578A (en) * | 2019-03-13 | 2019-07-12 | 五邑大学 | A kind of image rain removing method and its system, device, storage medium |
-
2019
- 2019-08-19 CN CN201910762822.6A patent/CN110458370A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101976501A (en) * | 2010-10-29 | 2011-02-16 | 天津市市政工程设计研究院 | Principal component analysis and neural network based port road safety prediction method |
| CN104599443A (en) * | 2015-01-12 | 2015-05-06 | 江苏省交通规划设计院股份有限公司 | Vehicle-mounted forewarning terminal for driving behaviors based on information fusion and forewarning method thereof |
| CN110009578A (en) * | 2019-03-13 | 2019-07-12 | 五邑大学 | A kind of image rain removing method and its system, device, storage medium |
Non-Patent Citations (1)
| Title |
|---|
| ALAMEEN NAJJAR等: "Combining Satellite Imagery and Open Data to Map Road Safety", 《PROCEEDINGS OF THE THIRTY-FIRST AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE (AAAI-17)》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111028512A (en) * | 2019-12-31 | 2020-04-17 | 福建工程学院 | Real-time traffic prediction method and device based on sparse BP neural network |
| CN111723997A (en) * | 2020-06-23 | 2020-09-29 | 上海电科智能系统股份有限公司 | Automatic generation method of urban major traffic accident data sample based on GAN |
| CN115018318A (en) * | 2022-06-01 | 2022-09-06 | 航天神舟智慧系统技术有限公司 | Method and system for predicting and analyzing risk of social area |
| CN115018318B (en) * | 2022-06-01 | 2023-04-18 | 航天神舟智慧系统技术有限公司 | Social area risk prediction analysis method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107967532B (en) | Urban traffic flow prediction method fusing regional vitality | |
| CN108710875B (en) | A kind of take photo by plane road vehicle method of counting and device based on deep learning | |
| Changzhen et al. | A traffic sign detection algorithm based on deep convolutional neural network | |
| CN109410575A (en) | A kind of road network trend prediction method based on capsule network and the long Memory Neural Networks in short-term of nested type | |
| CN105046956B (en) | Traffic flow simulating and predicting method based on turning probability | |
| CN109242291A (en) | River and lake basin water environment wisdom management method | |
| CN110458370A (en) | A kind of urban roads safety prediction technique and system based on open data | |
| CN106875424A (en) | A kind of urban environment driving vehicle Activity recognition method based on machine vision | |
| CN109410577A (en) | Adaptive traffic control sub-area division method based on Spatial Data Mining | |
| CN105355044B (en) | A kind of Expression method that urban road network lattice lock based on GIS is propagated | |
| CN106571030A (en) | Queuing length prediction method in multi-source traffic information environment | |
| CN112287832A (en) | High-resolution remote sensing image-based urban illegal building detection method | |
| CN109948471A (en) | Based on the traffic haze visibility detecting method for improving InceptionV4 network | |
| CN111160622A (en) | Scenic spot passenger flow prediction method and device based on hybrid neural network model | |
| CN109886147A (en) | A kind of more attribute detection methods of vehicle based on the study of single network multiple-task | |
| CN109242019A (en) | A kind of water surface optics Small object quickly detects and tracking | |
| CN110070720B (en) | Calculation method for improving fitting degree of traffic capacity model of intersection road occupation construction area | |
| CN115292913A (en) | Vehicle-road-cooperation-oriented drive test perception simulation system | |
| Lei et al. | Multitarget detection and tracking method in remote sensing satellite video | |
| Xu et al. | Building height calculation for an urban area based on street view images and deep learning | |
| CN104835142A (en) | Vehicle queuing length detection method based on texture features | |
| CN104331708B (en) | A kind of zebra crossing automatic detection analysis method and system | |
| CN115527118A (en) | Remote sensing image target detection method fused with attention mechanism | |
| Cui et al. | Skip Connection YOLO Architecture for Noise Barrier Defect Detection Using UAV-Based Images in High-Speed Railway | |
| CN101499214A (en) | Automatic traffic parameter extraction method based on image information entropy |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191115 |