CN110377670A - A kind of method, apparatus, medium and the equipment of determining road element information - Google Patents
A kind of method, apparatus, medium and the equipment of determining road element information Download PDFInfo
- Publication number
- CN110377670A CN110377670A CN201810322044.4A CN201810322044A CN110377670A CN 110377670 A CN110377670 A CN 110377670A CN 201810322044 A CN201810322044 A CN 201810322044A CN 110377670 A CN110377670 A CN 110377670A
- Authority
- CN
- China
- Prior art keywords
- road
- image
- road element
- candidate roads
- class
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/29—Geographical information databases
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
- G06V20/54—Surveillance or monitoring of activities, e.g. for recognising suspicious objects of traffic, e.g. cars on the road, trains or boats
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Remote Sensing (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to electronic map technique fields, method, apparatus, medium and the equipment of a kind of determining road element information are provided, in technical solution provided by the invention, the first kind road element of road image and the second class road element and the second class road element information of first kind road element information and road image are obtained, road element information includes the position of the confidence level and road element of road element in road image;Determine identical road element between first kind road element and the second class road element;If the identical road element is greater than first threshold in the confidence level of first kind road element information, and, confidence level in the second class road element information is less than second threshold, the identical road element is then determined as the road element information of the identical road element in the confidence level in first kind road element information and the position in the second class road element information.The present invention can obtain accurate road element information.
Description
Technical field
This application involves electronic map technique field more particularly to a kind of method, apparatus of determining road element information, it is situated between
Matter and equipment.
Background technique
With the development of road construction, electronic map data, which needs to update in real time very much, is just able to maintain electronic map data
Real-time and accuracy.Electronic map data includes the determination to road element information, currently, determining frequently with following manner
Road element information:
Image is acquired on road using image capture device interval set distance;Using acquired image as depth
The input for practising algorithm of target detection obtains the road element for including in each image of algorithm output and road element letter
Breath, wherein road element information includes the position of the confidence level and road element of road element in the picture.
At this stage, road element information mostly is determined using deep learning simple target detection algorithm, however utilize depth
Practising simple target detection algorithm, to obtain road element information not accurate enough, therefore, how to improve the accuracy of road element information
It is a problem in need of consideration.
Summary of the invention
The application provides method, apparatus, medium and the equipment of a kind of determining road element information, for solving the prior art
Present in using deep learning simple target detection algorithm obtain the not accurate enough problem of road element information.
On the one hand, the embodiment of the present application provides a kind of method of determining road element information, comprising:
For every determining road image, the first kind road of the road image determined using fast algorithm of detecting is obtained
Road element and first kind road element information, and, obtain the second class of the road image determined using accurate detection algorithm
Road element and the second class road element information, wherein road element information includes that the confidence level of road element and road are wanted
Position of the element in road image;Determine identical road between the first kind road element and the second class road element
Element;For each identical road element, if the confidence of the identical road element of this in the first kind road element information
Degree be greater than first threshold, also, in the second class road element information the identical road element confidence level less than second
Threshold value then wants the confidence level of the identical road element of this in the first kind road element information and the second class road
Position of the identical road element of this in prime information in the road image, is determined as the road element of the identical road element
Information, the first threshold are not less than the second threshold.In the embodiment of the present application, what is determined using fast algorithm of detecting
When road element with the road element determined using accurate detection algorithm is same road element, fast algorithm of detecting will be utilized
The confidence level for the same road element determined and by the road element determined using accurate detection algorithm in image
In road element information of the position as the same road element, to obtain accurate road element information.
In a kind of possible embodiment, road image is determined in the following way: according to each candidate roads image
Acquisition position, to the candidate roads image carry out hierarchical cluster, obtain for candidate roads image the first order classification knot
Fruit, wherein actual range in the classification of each first order between the adjacent candidate roads image of acquisition position no more than it is default away from
From;For obtained each first order classification, however, it is determined that in first order classification the adjacent candidate roads image of acquisition time it
Between road element it is similar, then the adjacent candidate roads image of the acquisition time is divided to same classification, obtains the second fraction
Class result;A candidate roads image is chosen from the classification of each second level respectively, as road image.This embodiment can
To greatly reduce the quantity for the road image for inputting accurate detection algorithm, and then it is true to improve accurate detection algorithm to a certain extent
Determine the processing speed of road element.
As a kind of possible embodiment, before determining road image, further includes: by collected each original image
Input as the fast algorithm of detecting;The original image comprising road element for determining fast algorithm of detecting output is
Candidate roads image.This embodiment can guarantee that all candidate roads images are the original image comprising road element,
In this way in the road element information for determining road image, it can avoid handling the road image for not including road element,
Improve the efficiency of determining road element information.
As a kind of possible embodiment, determine in first order classification the adjacent candidate roads image of acquisition time it
Between road element it is similar, specifically include: for the adjacent candidate roads image of acquisition time, it is forward to obtain acquisition time respectively
Candidate roads image the corresponding first hough transform frame of the first road element, and, the candidate roads of acquisition time rearward
The corresponding second hough transform frame of the second road element in image;If the first hough transform frame and second rectangle are examined
The deviation of the Aspect Ratio of survey frame is within preset range, the first hough transform frame is less than the second hough transform frame simultaneously
And the characteristics of image image district corresponding with the second hough transform frame of the corresponding image-region of the first hough transform frame
The characteristics of image in domain is similar, it is determined that the first road element is similar to the second road element.This embodiment knot
The deviation, hough transform frame size and hough transform frame for closing the Aspect Ratio of the hough transform frame of different road elements include
The characteristics of image of image-region determines similar road element, so as to make the accurate of the similar road element determined
Property is higher.
As a kind of possible embodiment, a candidate roads image is chosen from the classification of any second level, it is specific to wrap
It includes: for every candidate roads image in any second level classification, if each road element of the candidate roads image
Corresponding hough transform frame is completely contained in the candidate roads image, it is determined that the candidate roads image is complete mileage chart
Picture;According to image definition and picture size, a complete road image is chosen from determining complete road image.This reality
It applies mode combination image integrity, image definition and picture size and chooses candidate roads image from the classification of the second level, from
And it can guarantee and choose preferably candidate roads image relatively.
As a kind of possible embodiment, a complete road image, tool are chosen from determining complete road image
Body includes: the image definition and picture size that the complete road image is determined for every determining complete road image
Weighted sum is as a result, weight as the complete road image;The weight selection maximum one from determining complete road image
Open complete road image.
As a kind of possible embodiment, the weight of described image clarity is less than the weight of described image size.
As a kind of possible embodiment, it determines between the first kind road element and the second class road element
Identical road element, specifically includes: for each road element in first kind road element, determining the second class road element
In with the position of the road element in the road image have overlapping region road element, as be overlapped road element;If
The road element is greater than region threshold with the overlapping region for being overlapped road element, it is determined that the road element is overlapped road with this and wants
Element is identical road element.
As a kind of possible embodiment, the second class road element of the road image is determined using accurate detection algorithm
It with the second class road element information, specifically includes: intercepting the first kind road element from the road image, obtain the road
An image corresponding at least screenshot includes at least one road element in first kind road element in the screenshot;It will be each
Input of the Zhang Suoshu screenshot as the accurate detection algorithm obtains the road of the corresponding screenshot of the accurate detection algorithm output
Element and road element information;For every screenshot, by road element in the road element information of the screenshot in the screenshot
Position is mapped as position of the road element in the road image;Using the road element of each screenshot as the road image
Second class road element, and, the confidence level of the road element in the road element information of each screenshot and corresponding road are wanted
Position of the element in the road image is as the second class road element information.This possible embodiment, on the one hand ensure that
It include road element in every screenshot, the case where to avoid road element is not included in the screenshot for inputting accurate detection algorithm, from
And the processing load of accurate detection algorithm is reduced to a certain extent, on the other hand using screenshot as the defeated of accurate detection algorithm
Enter, reduce the size for inputting the image of accurate detection algorithm, is determined to improve accurate detection algorithm to a certain extent
The processing speed of road element information.
As a kind of possible embodiment, the first kind road element is intercepted from the road image, is specifically included:
Determine the first minimum circumscribed rectangle comprising all road elements in the first kind road element;If described first is minimum external
Rectangle is failed to grow up in length threshold or the width of first minimum circumscribed rectangle no more than width threshold value, then from the mileage chart
Interception includes the screenshot of all road elements in the first kind road element as in, and the length of the screenshot is not less than the length
Threshold value, also, the width of the screenshot is not less than the width threshold value.
As a kind of possible embodiment, if the length of first minimum circumscribed rectangle be greater than the length threshold or
First minimum circumscribed rectangle is wider than the width threshold value, then divides the road element in first kind road element
Group obtains multiple groupings comprising the length of the second minimum circumscribed rectangle of all road elements is equal to described in each grouping
Length threshold, also, the width of the second small boundary rectangle is equal to the width threshold value;For each grouping, from the mileage chart
Interception includes the screenshot of all road elements in the grouping as in.
On the other hand, the embodiment of the present application provides a kind of device of determining road element information, comprising:
Module is obtained, for obtaining the road determined using fast algorithm of detecting for every determining road image
The first kind road element and first kind road element information of image, and, obtain the road determined using accurate detection algorithm
Second class road element of road image and the second class road element information, wherein road element information includes setting for road element
The position of reliability and road element in road image;First determining module, for determine the first kind road element with
Identical road element between the second class road element;Second determining module, for being directed to each identical road element,
If the confidence level of the identical road element of this in the first kind road element information is greater than first threshold, also, described second
The confidence level of the identical road element of this in class road element information is less than second threshold, then believes the first kind road element
The identical road element is at this in the confidence level of the identical road element of this in breath and the second class road element information
Position in road image is determined as the road element information of the identical road element, and the first threshold is not less than described
Second threshold.
As a kind of possible embodiment, the device of determining road element information provided by the embodiments of the present application, is also wrapped
It includes: third determining module, for determining road image in the following way: according to the acquisition position of each candidate roads image,
Hierarchical cluster is carried out to the candidate roads image, obtains the first order classification results for candidate roads image, wherein is each
Actual range in first order classification between the adjacent candidate roads image of acquisition position is not more than pre-determined distance;For what is obtained
Each first order classification, however, it is determined that the road element phase in first order classification between the adjacent candidate roads image of acquisition time
Seemingly, then the adjacent candidate roads image of the acquisition time is divided to same classification, obtains second level classification results;Respectively from every
A candidate roads image is chosen in a second level classification, as road image.
As a kind of possible embodiment, the device of determining road element information provided by the embodiments of the present application, is also wrapped
It includes:
4th determining module, for before the third determining module determines road image, by collected each original
Input of the beginning image as the fast algorithm of detecting;Determine the original comprising road element of the fast algorithm of detecting output
Image is candidate roads image.
As a kind of possible embodiment, the third determining module, is specifically used for: for the adjacent time of acquisition time
Road image is selected, obtains corresponding first hough transform of the first road element of the forward candidate roads image of acquisition time respectively
Frame, and, the corresponding second hough transform frame of the second road element in the candidate roads image of acquisition time rearward;If described
The deviation of the Aspect Ratio of first hough transform frame and the second hough transform frame is within preset range, first rectangle
Detection block be less than the second hough transform frame and the characteristics of image of the corresponding image-region of the first hough transform frame with
The characteristics of image of the corresponding image-region of the second hough transform frame is similar, it is determined that the first road element and described the
Two road elements are similar.
As a kind of possible embodiment, the third determining module is specifically used in the following way from any second
A candidate roads image is chosen in grade classification: for every candidate roads image in the classification of any second level, if the candidate
The corresponding hough transform frame of each road element of road image is completely contained in the candidate roads image, it is determined that the time
Selecting road image is complete road image;According to image definition and picture size, chosen from determining complete road image
One complete road image.
As a kind of possible embodiment, the third determining module is specifically used in the following way from determining complete
A complete road image is chosen in whole road image: for determining every complete road image, determining the complete mileage chart
The weighted sum of the image definition and picture size of picture is as a result, weight as the complete road image;From determining complete
The maximum complete road image of weight selection in road image.
As a kind of possible embodiment, the weight of described image clarity is less than the weight of described image size.
As a kind of possible embodiment, first determining module, is specifically used for: in first kind road element
Each road element, determine in the second class road element with the position of the road element in the road image have be overlapped area
The road element in domain, as coincidence road element;If the road element is greater than region threshold with the overlapping region for being overlapped road element
Value, it is determined that it is identical road element that the road element, which is overlapped road element with this,.
As a kind of possible embodiment, the device of determining road element information provided by the embodiments of the present application, is also wrapped
It includes: the 5th determining module, for realizing the second class road for determining the road image using accurate detection algorithm in the following way
Road element and the second class road element information: the first kind road element is intercepted from the road image, obtains the mileage chart
It include at least one road element in first kind road element as a corresponding at least screenshot, in the screenshot;By each
Input of the screenshot as the accurate detection algorithm, the road for obtaining the corresponding screenshot of the accurate detection algorithm output are wanted
Element and road element information;For every screenshot, by position of the road element in the screenshot in the road element information of the screenshot
It sets, is mapped as position of the road element in the road image;Using the road element of each screenshot as the of the road image
Two class road elements, and, by the confidence level of the road element in the road element information of each screenshot and corresponding road element
Position in the road image is as the second class road element information.
As a kind of possible embodiment, the 5th determining module is specifically used in the following way from the mileage chart
The first kind road element is intercepted as in: determining that first comprising all road elements in the first kind road element is minimum
Boundary rectangle;If first minimum circumscribed rectangle is failed to grow up in the width of length threshold or first minimum circumscribed rectangle
It then include section of all road elements in the first kind road element from interception in the road image no more than width threshold value
Figure, the length of the screenshot is not less than the length threshold, also, the width of the screenshot is not less than the width threshold value.
As a kind of possible embodiment, the 5th determining module is also used to: if first minimum circumscribed rectangle
Length be greater than the length threshold or first minimum circumscribed rectangle be wider than the width threshold value, then to first kind road
Road element in the element of road is grouped, obtain it is multiple grouping comprising in each grouping all road elements second
The length of minimum circumscribed rectangle is equal to the length threshold, also, the width of the second small boundary rectangle is equal to the width threshold value;
It include the screenshot of all road elements in the grouping from interception in the road image for each grouping.
Another aspect, the embodiment of the present application provide a kind of nonvolatile computer storage media, and the computer storage is situated between
Matter is stored with executable program, which, which executes, realizes that any determination road provided by the above embodiment is wanted
The step of method of prime information.
In another aspect, the embodiment of the present application provides a kind of computer equipment, including memory, processor and it is stored in storage
Computer program on device, the processor realize any determination road element provided by the above embodiment when executing described program
The step of method of information.
Method, apparatus, medium and the equipment for the determination road element information that the embodiment of the present application mentions, fast algorithm of detecting and
Accurate detection algorithm belongs to deep learning multi-target detection algorithm, considers the road element determined using fast algorithm of detecting
The high these two aspects of positional accuracy of the road element that confidence level is relatively reliable and the accurate detection algorithm of utilization is determining in the picture
The advantages of, it is with the road element determined using accurate detection algorithm in the road element determined using fast algorithm of detecting
When same road element, by the confidence level for the same road element determined using fast algorithm of detecting and it will utilize accurate
The road element information of the position of road element in the picture as the same road element that detection algorithm is determined, thus
Obtain accurate road element information.
Detailed description of the invention
Fig. 1 is application scenarios schematic diagram provided by the embodiments of the present application;
Fig. 2 is the method flow schematic diagram of determining road element information provided by the embodiments of the present application;
Fig. 3 is the overlapping region schematic diagram of two road elements provided by the embodiments of the present application in the picture;
Fig. 4 is the method flow schematic diagram of determining road image provided by the embodiments of the present application;
Road element similar method flow of the Fig. 5 between determining candidate roads image provided by the embodiments of the present application shows
It is intended to;
Fig. 6 is the method flow signal provided by the embodiments of the present application that candidate roads image is chosen from the classification of the second level
Figure;
Fig. 7 is the method flow schematic diagram provided by the embodiments of the present application for choosing complete road image;
Fig. 8 is the second class road element and the second class road element letter of determining road image provided by the embodiments of the present application
The method flow schematic diagram of breath;
Fig. 9 a is position view of the road element in screenshot in screenshot provided by the embodiments of the present application;
Fig. 9 b is position view of the screenshot provided by the embodiments of the present application in road image;
Fig. 9 c is position view of the road element in road image in screenshot provided by the embodiments of the present application;
Figure 10 is the method flow signal provided by the embodiments of the present application that first kind road element is intercepted from road image
Figure;
Figure 11 is the method flow schematic diagram provided by the embodiments of the present application being grouped to road element;
Figure 12 is the apparatus structure schematic diagram of determining road element information provided by the embodiments of the present application;
Figure 13 is provided by the embodiments of the present application for determining that the hardware configuration of the computer equipment of road element information shows
It is intended to.
Specific embodiment
To keep the purposes, technical schemes and advantages of the application clearer, below in conjunction with attached drawing to the application embodiment party
Formula is described in further detail.
Referenced herein " multiple " refer to two or more."and/or", the association for describing affiliated partner are closed
System indicates may exist three kinds of relationships, for example, A and/or B, can indicate: individualism A exists simultaneously A and B, individualism
These three situations of B.Character "/" typicallys represent the relationship that forward-backward correlation object is a kind of "or".
Herein, it is to be understood that any number of elements in attached drawing be used to example rather than limit and it is any
Name is only used for distinguishing, without any restrictions meaning.In order to facilitate understanding, below to involved in the embodiment of the present application
Noun explains.
Road element: playing the information of mark action in road, for example may include traffic lights, traffic signboard, terrestrial reference
Line, zebra stripes etc..
Road element information: the position of confidence level and road element in the picture including road element, wherein road
The credibility for the road element that the confidence level of element is used to indicate to determine, for example the confidence level of traffic lights is this determined
Road element is the credibility of traffic lights;The position of road element in the picture is the corresponding rectangle of road element determined
The position of detection block in the picture.
Fast algorithm of detecting: belonging to deep learning multi-target detection algorithm, can detect that using the algorithm more in image
A target simultaneously can determine that the position of the confidence level and each target of each target in the picture, detects fast speed, determines
Target confidence level it is relatively reliable, but the accuracy of the position of the target determined in the picture is poor, common quickly detection
Algorithm includes yolo v1 (the yolo algorithm of first version), yolo v2 (the yolo algorithm of the second edition) etc., wherein yolo is
The abbreviation of You Only Look Once, yolo algorithm are unified, real-time algorithm of target detection (Unified, Real-Time
Object Detection), yolo v2 ratio yolo v1 is more preferable, faster, it is stronger, fast algorithm of detecting is applied to the present invention and implements
When example, target that road element as needs to detect.
Accurate detection algorithm: belonging to deep learning multi-target detection algorithm, can detect that using the algorithm more in image
A target simultaneously can determine that the position of the confidence level and each target of each target in the picture, and the target determined is in the picture
Position accuracy it is higher, but it is slower to detect speed, common accurate detection algorithm include: R-CNN, Faster R-CNN,
SSD etc., wherein the full name in English of R-CNN are as follows: Regions with convolutionalneural networks
The Chinese name of features, R-CNN are the precision target detection method based on region convolutional neural networks feature, Faster
The full name in English of R-CNN are as follows: Faster Regions with convolutionalneural networks features,
The Chinese name of Faster R-CNN are as follows: the precision target detection method quickly based on region convolutional neural networks feature, SSD
Full name in English are as follows: the Chinese name of Single Shot MultiBox Detector, SSD are as follows: single deep neural network is more
Frame object detector, when accurate detection algorithm is applied to the embodiment of the present invention, target that road element as needs to detect.
Characteristics of image: the feature of image itself, such as color histogram or image texture characteristic for image.
Currently, often determining the road element and road element of image using fast algorithm of detecting or accurate detection algorithm
Information, inventors have found that the accuracy of the position of the road element obtained using fast algorithm of detecting in the picture is poor, and it is sharp
The confidence level of the road element determined with accurate detection algorithm is than the confidence level of the road element determined using fast algorithm of detecting
Poor reliability, this can all lead to the problem of road element information inaccuracy, and road element information inaccuracy can directly result in obtain
Map datum it is unreliable, therefore, it is necessary to a kind of schemes of improved determining road element information, are wanted with obtaining accurate road
Prime information.
For this purpose, the embodiment of the present application provides a kind of method of determining road element information, this method may include: for true
Every fixed road image obtains the first kind road element and the first kind of the road image determined using fast algorithm of detecting
Road element information, and, obtain the second class road element and second of the road image determined using accurate detection algorithm
Class road element information, wherein the confidence level and road element that road element information includes road element are in road image
Position;Determine identical road element between first kind road element and the second class road element;For each identical road
Road element, if the confidence level of the identical road element of this in first kind road element information is greater than first threshold, also, the second class
The confidence level of the identical road element of this in road element information is less than second threshold, then will be somebody's turn to do in first kind road element information
The identical road element is in the road image in the confidence level of identical road element and the second class road element information
Position, be determined as the road element information of the identical road element, wherein first threshold be not less than second threshold.
The method of determining road element information provided by the embodiments of the present application considers the road determined using fast algorithm of detecting
The confidence level of road element is relatively reliable and high using the positional accuracy of the determining road element of accurate detection algorithm in the picture
The advantages of these two aspects, in the road element determined using fast algorithm of detecting and the road determined using accurate detection algorithm
When road element is same road element, by the confidence level for the same road element determined using fast algorithm of detecting and incite somebody to action
The position of road element in the picture determined using accurate detection algorithm is as the road element of the same road element
Information, to obtain accurate road element information.
Below with reference to the application scenarios that Fig. 1 is provided, to the scheme of determining road element information provided by the embodiments of the present application
It is illustrated.
As shown in Figure 1, including database server 101, calculation server 102.Wherein, it is deposited in database server 101
The first kind road element determined using fast algorithm of detecting and first kind road element information have been stored up, and, utilize accurate inspection
The the second class road element and the second class road element information that method of determining and calculating determines.Calculation server 102 is from database server 101
The first kind road element and first kind road element information of road image are obtained, and, the second class road of the road image
Element and the second class road element information;Determine that identical road is wanted between first kind road element and the second class road element
Element;For each identical road element, if the confidence level of the identical road element of this in first kind road element information is greater than
First threshold, also, the confidence level of the identical road element is less than second threshold in the second class road element information, then by the
The identical road in the confidence level of the identical road element of this in a kind of road element information and the second class road element information
Position of the road element in the road image is determined as the road element information of the identical road element, wherein first threshold
Not less than second threshold.The road element information of determining same link element can be sent to client by calculation server 102
103 are shown, related technical personnel can check the road element information of determining same link element by client 103.
In Fig. 1, database server 101 and calculation server 102 can pass through local area network, wide area network or mobile Internet etc.
Communication network is communicated, and calculation server 102 and client 103 can be logical by local area network, wide area network or mobile Internet etc.
Communication network is communicated.Database server 101, server apparatus 102 and client 103 can for portable equipment (such as:
Mobile phone, plate, laptop etc.), or PC (PC, Personal Computer).
As another application scenarios, database server and calculation server in Fig. 1 can set for same calculating
Standby, as another application scenarios, database server and calculation server in Fig. 1 can be same calculating equipment, and
And the calculating equipment is the calculating equipment for including display screen, related technical personnel can be looked by the display screen of the calculating equipment at this time
It sees the road element information of determining road element, i.e., may not include individual client in the application scenarios, but calculating
The function of the display road element information of integrated client in equipment.
It should be noted that foregoing relate to application scenarios be merely for convenience of understanding spirit herein and principle and show
Out, the embodiment of the present application is unrestricted in this regard.On the contrary, the embodiment of the present application can be applied to applicable any field
Scape.
Below with reference to Fig. 2, the method for determining road element information provided by the embodiments of the present application is illustrated.
As shown in Fig. 2, the method for determining road element information provided by the embodiments of the present application, may comprise steps of:
Step 201, for every determining road image, the road image determined using fast algorithm of detecting is obtained
First kind road element and first kind road element information, and, obtain the road image determined using accurate detection algorithm
The second class road element and the second class road element information, wherein road element information include the confidence level of road element with
And position of the road element in road image.
When it is implemented, advancing with fast algorithm of detecting and accurate detection algorithm determines the road element of road image
With road element information, wherein the road element of the road image determined using fast algorithm of detecting is known as first kind road
The road element information of the road image determined using fast algorithm of detecting is known as first kind road element information by element, will
The road element of the road image determined using accurate detection algorithm is known as the second class road element, will utilize accurate detection algorithm
The road element information of determining road image is known as the second class road element information.It include at least one in first kind road element
A road element includes the confidence level of each road element and each in first kind road element in first kind road element information
Position of a road element in road image;It include at least one road element, the second class road in second class road element
Confidence level and each road element including each road element in the second class road element in element information is in road image
In position.
The first kind road element and first kind road element information of road image are determined using fast algorithm of detecting
Mode is as follows: training quick detection model in advance;Diminution processing is carried out to road image;Using the road image after diminution as pre-
The first input of trained quick detection model, obtains the output of quick detection model, carries out to the output of quick detection model
Non- maximum suppression processing, obtains the first kind road element and first kind road element information of road image, wherein to mileage chart
As carrying out reducing processing being specially to be contracted to road image to guarantee the precision that multiple tracks road element is recalled.
The the second class road element and the second class road element information of road image are determined using accurate detection algorithm
Mode is as follows: training accurate detection model in advance;Using road image as the input of preparatory trained accurate detection model, obtain
To the output of accurate detection model, fusion is carried out to the output of accurate detection model and non-maximum suppression is handled, obtains mileage chart
Second class road element of picture and the second class road element information.
It should be noted that road image is to be acquired at the different location of the predeterminable area of road using acquisition equipment
Image, for example acquisition equipment is installed to collecting vehicle, the image in front of collecting vehicle is acquired in collecting vehicle moving process.
Step 202, identical road element between first kind road element and the second class road element is determined.
When it is implemented, first kind road element and the second class road element are to be determined using different detection algorithms
The road element of same road image, therefore, there are identical roads between first kind road element and the second class road element
A possibility that element, is larger, for example, the road element of road image includes traffic lights 1, it include red green in first kind road element
It also include traffic lights 1 in lamp 1 and the second class road element, then it is assumed that traffic lights 1 are first kind road element and the second class road
Identical road element between the element of road.
In practical application, it may be determined that the characteristics of image of each road element and the second class road in first kind road element
The similar two road elements of characteristics of image are determined as same link element by the characteristics of image of each road element in element,
Wherein, in the similar two road elements of characteristics of image, one of road element belongs to first kind road element, another road
Road element belongs to the second class road element.Can also determine otherwise first kind road element and the second class road element it
Between identical road element, here without limitation.
Optionally, determine that identical road is wanted between first kind road element and the second class road element in the following way
Element:
For each road element in first kind road element, determines in the second class road element and exist with the road element
Position in the road image has the road element of overlapping region, as coincidence road element;If the road element be overlapped
The overlapping region of road element is greater than region threshold, it is determined that it is that identical road is wanted that the road element, which is overlapped road element with this,
Element.
When it is implemented, road element A can be denoted as the road element in first kind road element, the second class road is wanted
Road element in element is denoted as road element B, for each road element A, determines and wants in the second class road element with the road
Position of the plain A in road image has the road element B of overlapping region, as road element is overlapped, that is, is overlapped road element
Belong to the second class road element.Wherein, overlapping region is that the corresponding hough transform frame of road element A and road element B are corresponding
Overlapping region of the hough transform frame in road image.Region threshold can be equal to the corresponding hough transform frame area of road element
Setting ratio, for example be the 80% of the corresponding hough transform frame area of road element, wherein the road element can be the first kind
Road element may be the second class road element.
When it is implemented, if the coincidence of position of the corresponding hough transform frame of two road elements in same road image
Region is greater than region threshold, it is determined that the two road elements are identical road element, wherein one in two road elements
A road element (road element A) belongs to first kind road element, another road element (road element B) belongs to the second class road
Road element.As shown in figure 3, position of the road element A in road image in first kind road element is wanted with the second class road
The overlapping region of the position of road element B in the picture in element is region C, at this point, if region C is greater than region threshold, really
Determining road element A and road element B is identical road element, for example road element A and road element B are traffic lights 1, no
Then, determine that road element A and road element B are different road elements, for example road element A is traffic lights 1, road element B
For direction board 1.
Step 203, for each identical road element, if the identical road element of this in first kind road element information
Confidence level be greater than first threshold, also, in the second class road element information the identical road element confidence level less than the
Two threshold values, then by the confidence level of the identical road element of this in first kind road element information and the second class road element information
In the position of the identical road element in the road image, be determined as the road element information of the identical road element,
First threshold is not less than second threshold.
When it is implemented, determine between first kind road element and the second class road element identical road element it
Afterwards, for each identical road element, if the confidence level of the identical road element of this in first kind road element information is greater than
The confidence level of the identical road element is less than second threshold in first threshold and the second class road element, then by first kind road
The identical road element in the confidence level of the identical road element of this in the element information of road and the second class road element information
Position in the road image is determined as the road element information of the identical road element.
Specifically, it is not less than the road element of second threshold for the confidence level of road element in the second class road element,
The road element information for determining the road element is the road element information of the road element in the second class road element information.Needle
To road element different between first kind road element and the second class road element, if the different road element belongs to first
Class road element, and the confidence level of the different road element of this in first kind road element information is not more than first threshold, then
Still by the confidence level of the different road element of this in first kind road element information and the road element in road image
The position road element different as this road element information.
The method of determining road element information provided by the embodiments of the present application, fast algorithm of detecting and accurate detection algorithm are all
Belong to deep learning multi-target detection algorithm, consider the confidence level of the road element determined using fast algorithm of detecting it is relatively reliable with
And the advantages of high using road element positional accuracy in the picture that accurate detection algorithm determines these two aspects, using fast
When road element that fast detection algorithm is determined with the road element determined using accurate detection algorithm is same road element,
It will be determined the confidence level for the same road element determined using fast algorithm of detecting and using accurate detection algorithm
The road element information of the position of road element in the picture as the same road element, to obtain accurate road
Road element information.
In practical application, it can be trained according to sample data training classifier, to determine the unknown parameter in classifier,
By the confidence level of road element A and road element A position in the picture, the confidence level and road element B of road element B
Position in the picture as input parameter, unknown input parameter determine after classifier, if the output result of classifier be it is yes,
Then will be considered to road element A and road element B is identical road element, and the confidence level of road element A and road are wanted
Position of the plain B in the road image is determined as the road element information of the identical road element.Wherein, road element A belongs to
In first kind road element, road element B belongs to the second class road element.Using the position of road element in the picture as input
Parameter is specially using the position in the picture of four vertex in the corresponding hough transform frame of road element as inputting parameter.Into
One step, input parameter can also include the type of road element, i.e. road element A belongs to first kind road element, and road is wanted
Plain B belongs to the second class road element.It is trained according to sample data training classifier, to determine the unknown parameter in classifier
Process can be found in the prior art, be not described herein.
As a kind of possible embodiment, road image can be determined according to the content that Fig. 4 is provided:
Step 401, according to the acquisition position of each candidate roads image, hierarchical cluster is carried out to candidate roads image, is obtained
To the first order classification results for being directed to candidate roads image, wherein the adjacent candidate road of acquisition position in each first order classification
Actual range between the image of road is not more than pre-determined distance.
When it is implemented, candidate roads image can be the acquisition collected original image of equipment, or to from adopting
The image that filters out in the collection collected original image of equipment, for example, from that will obscure and the image of backlight is from original image
It deletes, using remaining original image as candidate roads image, certainly, candidate roads figure can also be filtered out otherwise
Picture, here without limitation.
In this step, the acquisition position of each candidate roads image is specially to acquire to set when acquiring corresponding candidate road image
Standby locating geographical location, the GPS of acquisition equipment when which can be from the acquisition corresponding candidate road image of record
It is extracted in (Global Positioning System, global positioning system) information, geographical location specifically may include longitude and latitude
Degree.
In this step, candidate roads image is clustered using hierarchical clustering algorithm, specifically using the layering of cohesion
Clustering algorithm clusters candidate roads image, can refer to the prior art using the process that hierarchical clustering algorithm is clustered,
Here it is not detailed.
After carrying out hierarchical cluster to candidate roads image, obtained first order classification results include at least one first fraction
Class, the actual range in the classification of each first order between the adjacent candidate roads image of acquisition position are not more than pre-determined distance, in advance
If distance can be arranged according to practical application scene, for example may be configured as 30 meters, 50 meters etc., here without limitation.
Each candidate roads image in each first order classification is ranked up according to acquisition position, wherein assuming that first
Between the acquisition position of the candidate roads image of the acquisition position and the second specified sequence of the candidate roads image of specified sequence
Actual range is the first actual range;The acquisition position and third of the candidate roads image of second specified sequence specify the time of sequence
Selecting actual range between the acquisition position of road image is the second actual range, then the first actual range it is practical less than second away from
From;If other candidates are not present between the candidate roads image of the first specified sequence and the candidate roads image of the second specified sequence
Road image, then it is assumed that the candidate roads image of the candidate roads image of the first sequence and the second specified sequence is acquisition position phase
Adjacent candidate roads image;Correspondingly, if the candidate roads image of the second specified sequence and third specify the candidate roads of sequence
Other candidate roads images are not present between image, then it is assumed that the candidate roads image and third of the second sequence specify the time of sequence
Selecting road image is the adjacent candidate roads image of acquisition position.Corresponding sort of second specified sequence is located at the first specified sequence
Between corresponding sequence and the corresponding sequence of third execution sequence.
For example, including candidate roads image 1, candidate roads image 2, candidate roads image 3 and time in first order classification
Select road image 4, wherein the result being ranked up according to acquisition position is candidate roads image 2, candidate roads image 1, candidate
Road image 3 and candidate roads image 4, then, in first order classification, candidate roads image 2 and candidate roads image 1
Acquisition position is adjacent, candidate roads image 1 is adjacent with the acquisition position of candidate roads image 3, candidate roads image 3 and candidate road
The acquisition position of road image 4 is adjacent, wherein between the acquisition position of the adjacent any two candidate roads images of acquisition position
Actual range is not more than pre-determined distance.Actual range between the non-conterminous two candidate roads images of acquisition position is greater than acquisition
The distance between two temporally adjacent candidate images, such as the acquisition position and candidate roads figure of above-mentioned candidate roads image 2
Actual range between the acquisition position of picture 1 is less than the acquisition position of candidate roads image 1 and the acquisition position of candidate roads image 3
Actual range between setting;Between the acquisition position of candidate roads image 1 and the acquisition position of candidate roads image 3 it is practical away from
The actual range between acquisition position from the acquisition position and candidate roads image 4 that are less than candidate roads image 3.
Step 402, classify for obtained each first order, however, it is determined that the adjacent time of acquisition time in first order classification
It selects the road element between road image similar, then the adjacent candidate roads image of the acquisition time is divided to same classification,
Obtain second level classification results.
When it is implemented, after obtaining first order classification results, further directed to the candidate roads in the classification of each first order
Image is classified, and second level classification results are obtained.Wherein, according to the road between the adjacent candidate roads image of acquisition time
The similitude of element carries out subseries again to the candidate roads image in first order classification.Specifically to the time in first order classification
, can be according to acquisition time by the early sequence to evening when road image being selected to be classified again, circulation executes following steps, until right
A candidate roads image of the acquisition time after completes classification in first order classification:
Determine whether is road element between the candidate roads image of current order and the candidate roads image of next sequence
It is similar;If so, the candidate roads image of the candidate roads image of current order and next sequence is divided to same classification;It will
The candidate roads image of next sequence as current order candidate roads image and execute the candidate roads of determining current order
The whether similar step of road element between image and the candidate roads image of next sequence;If it is not, by the time of current order
It selects road image and the candidate roads image of next sequence to be divided to different classifications, and the candidate roads image of next sequence is made
For current order candidate roads image and execute the candidate roads image of determining current order and the candidate roads of next sequence
The whether similar step of road element between image.
Wherein, can determine between the adjacent candidate roads image of acquisition time same type of road element it is similar or
Determine that any road element is similar or is determining the adjacent time of acquisition time between the adjacent candidate roads image of acquisition time
It selects in the case that similar road element number reaches certain amount between road image, by the adjacent candidate roads of acquisition time
Image is divided to same classification.Certain value can be greater than in the similarity of the characteristics of image of the corresponding image-region of two road elements
In the case where, determine that two road elements are similar, alternatively, can also determine that two road elements are similar otherwise, here
Without limitation.
Optionally, before executing step 402, further screening belongs to the candidate roads figure of ascending pathway in same road
The candidate roads image of picture and downlink road, the first order can specifically be classified in acquisition time be spaced in candidate in preset duration
Road image is as first group, for example it is believed that all candidate roads images in first group are the candidate road on ascending pathway
Road image.For the candidate roads image in first group of obtained each first order classification, however, it is determined that acquisition time in the group
Road element between adjacent candidate roads image is similar, then is divided to the adjacent candidate roads image of the acquisition time together
One classification, obtains second level classification results.
By the first order classify in candidate roads image except first group be divided into second group, such as it is believed that in second group
All candidate roads images be candidate roads image on downlink road, and the candidate roads image in second group is carried out
Second level classification, alternatively, since the acquisition time interval between the candidate roads image in second group may not be in preset duration
It is interior, therefore the unallocated candidate roads image in first group as undesirable image and can be abandoned.
Step 403, a candidate roads image is chosen from the classification of each second level respectively, as road image.
When it is implemented, obtained second level classification results include the classification of at least one second level, each second level classification
In include an at least candidate roads image, respectively from each second level classification in choose a candidate roads image as road
Image, so that determining road image is obtained, at this point, the quantity of the road image determined is identical as the number that the second level is classified.
A candidate roads image is randomly selected in can classifying from each second level, alternatively, according to other selection rules from the second fraction
A road image is chosen in class, here without limitation.It in practical applications, can also be according to practical application scene, respectively from each
Multiple candidate roads images are chosen in a second level classification, for example choose two candidate roads from the classification of each second level respectively
Image, as road image.
In this embodiment, the classification in acquisition position is first carried out to candidate roads image and obtains first order classification knot
Fruit carries out second level classification to each first order classification results further according to the similitude of road element, at this time it is believed that obtaining
The similitude between candidate roads image in each second level classification is higher, chooses one from the classification of each second level respectively
Image is as road image, so as to greatly reduce the quantity for the road image for inputting accurate detection algorithm, and then certain
The processing speed that accurate detection algorithm determines road element is improved in degree.
Fig. 4 is only a kind of embodiment of possible determining road image, not to determining mileage chart in the embodiment of the present application
The real-time mode of picture is defined, for example can also determine in the following way road image:
Mode one: according to the acquisition position of each candidate roads image, hierarchical cluster is carried out to candidate roads image, is obtained
For the classification results of candidate roads image, an at least candidate roads image is chosen from each classification respectively as mileage chart
Picture.
Mode two: if it is determined that the road in each candidate roads image between the adjacent candidate roads image of acquisition time is wanted
Element is similar, then the adjacent candidate roads image of the acquisition time is divided to same classification, obtains for candidate roads image
Classification results choose an at least candidate roads image as road image from each classification respectively.
As a kind of possible embodiment, before determining road image, determining road provided by the embodiments of the present application
The method of element information, further comprising the steps of:
Using collected each original image as the input of fast algorithm of detecting;Determine the packet of fast algorithm of detecting output
The original image of the element containing road is candidate roads image.
When it is implemented, may not include road element in acquisition equipment acquired image, in order to avoid will not include
The image of road element determines the original image comprising road element as road image, using fast algorithm of detecting, from
And can guarantee that all candidate roads images are the original image comprising road element, in this way in the road for determining road image
When element information, it can avoid handling the road image for not including road element, improve determining road element information
Efficiency.
As a kind of possible embodiment, the content that can be provided according to Fig. 5 determines acquisition time in first order classification
Road element between adjacent candidate roads image is similar:
Step 501, the candidate roads image adjacent for acquisition time obtains the forward candidate roads of acquisition time respectively
The corresponding first hough transform frame of the first road element of image, and, in the candidate roads image of acquisition time rearward
The corresponding second hough transform frame of two road elements.
When it is implemented, the adjacent candidate roads image of acquisition time includes two candidate roads images, wherein a time
Select the acquisition time of road image more forward than the acquisition time of another candidate roads image.First road element is acquisition time
Any road element in forward candidate roads image, the second road element are in the candidate roads image of acquisition time rearward
Any road element.
Wherein, position of the first hough transform frame in road image is position of the first road element in road image
It sets, position of the second hough transform frame in road image is position of the second road element in road image.
Step 502, if the deviation of the Aspect Ratio of the first hough transform frame and the second hough transform frame preset range it
Interior, the first hough transform frame is less than the second hough transform frame and the characteristics of image of the corresponding image-region of the first hough transform frame
The characteristics of image of image-region corresponding with the second hough transform frame is similar, it is determined that the first road element and the second road element
It is similar.
When it is implemented, determining between the Aspect Ratio of the first hough transform frame and the Aspect Ratio of the second hough transform frame
Deviation, for example, by the difference between the Aspect Ratio of the first hough transform frame and the Aspect Ratio of the second hough transform frame
Deviation of the absolute value as the two, alternatively, by the length-width ratio of the Aspect Ratio of the first hough transform frame and the second hough transform frame
Deviation of the difference as the two between example, alternatively, by the Aspect Ratio of the first hough transform frame and the second hough transform frame
Deviation of the ratio of Aspect Ratio as the two.The size of preset range can be set according to practical application scene, not limited here
It is fixed.
When it is implemented, the first hough transform frame area less than the second hough transform frame area when, determine, first
Hough transform frame is less than the second hough transform frame.If including that the higher road of similitude is wanted in collected candidate roads image
Element, since acquisition equipment is to acquire candidate roads image according to principle from the distant to the near, wherein far adopted apart from road element
The acquisition time of the image collected is more forward, therefore the corresponding rectangle of road element in the forward candidate roads image of acquisition time
Detection block hough transform frame more corresponding than the road element in acquisition time candidate roads image rearward is small.
When it is implemented, the corresponding hough transform frame of road element is shown on candidate roads image, hough transform frame pair
The image-region answered is specially the image-region on the candidate roads image that hough transform frame includes.It is each in the embodiment of the present application
The characteristics of image of hough transform frame is same type of characteristics of image, for example is the color histogram of image or is figure
As textural characteristics.
This possible embodiment is intended to illustrate to determine that two road elements between different candidate roads images are similar
Process.Deviation, rectangle in this possible embodiment, in conjunction with the Aspect Ratio of the hough transform frame of different road elements
The characteristics of image for the image-region that detection block size and hough transform frame include determines similar road element, so as to
Keep the accuracy for the similar road element determined higher.
Certainly, the content that Fig. 5 is provided is only a kind of whether similar embodiment of possible determining road element, can also be tied
The deviation, hough transform frame size and hough transform frame for closing the Aspect Ratio of the hough transform frame of different road elements include
At least two in the characteristics of image of image-region, determine whether road element is similar, such as the adjacent time of acquisition time
Road image is selected, obtains corresponding first hough transform of the first road element of the forward candidate roads image of acquisition time respectively
Frame, and, the corresponding second hough transform frame of the second road element in the candidate roads image of acquisition time rearward;If first
The deviation of the Aspect Ratio of hough transform frame and the second hough transform frame is within preset range and the first hough transform frame is small
In the second hough transform frame, it is determined that the first road element is similar to the second road element;It is adjacent for acquisition time for another example
Candidate roads image, obtain corresponding first rectangle of the first road element of the forward candidate roads image of acquisition time respectively
Detection block, and, the corresponding second hough transform frame of the second road element in the candidate roads image of acquisition time rearward;If
The deviation of the Aspect Ratio of first hough transform frame and the second hough transform frame is within preset range and the first hough transform
The characteristics of image of the characteristics of image of the corresponding image-region of frame image-region corresponding with the second hough transform frame is similar, it is determined that
First road element is similar to the second road element.
As a kind of possible embodiment, the content that can be provided according to Fig. 6 chooses one from the classification of any second level
Candidate roads image:
Step 601, for every candidate roads image in the classification of any second level, if the candidate roads image is each
The corresponding hough transform frame of road element is completely contained in the candidate roads image, it is determined that the candidate roads image has been
Whole road image.
When it is implemented, may include incomplete hough transform frame in candidate roads image, illustrate the candidate road at this time
It include incomplete road element in the image of road, i.e. the integrality of the candidate roads image is not good enough, guarantees in this step
The integrality of image, each road element in candidate roads image correspond to hough transform frame and are completely contained in the candidate road
When in the image of road, determine that the candidate roads image is complete road image.
Step 602, according to image definition and picture size, a full track is chosen from determining complete road image
Road image.
When it is implemented, after the integrality for guaranteeing image, according to the clarity and picture size of image, from determining complete
A complete road image is chosen in whole road image.Specifically, one can be chosen from determining complete road image clearly
Degree is greater than clarity threshold and picture size is greater than the complete road image of size threshold.
In this possible embodiment, in conjunction with image integrity, image definition and picture size from the second fraction
Candidate roads image is chosen in class, chooses preferably candidate roads image relatively to can guarantee.
Fig. 6 is only a kind of possible embodiment realized and choose candidate roads image from the classification of the second level, in reality
In, it is contemplated that at least one of image integrity, image definition and picture size three can be from the second fractions
Candidate roads image is chosen in class, for example, choosing the preferable candidate roads image of a clarity from the classification of the second level.
As a kind of possible embodiment, the content that can be provided according to Fig. 7, is realized according to image definition and image ruler
It is very little, a complete road image is chosen from determining complete road image:
Step 701, for determining every complete road image, the image definition and figure of the complete road image are determined
As the weighted sum of size is as a result, weight as the complete road image.
Step 702, the maximum complete road image of weight selection from determining complete road image.
When it is implemented, the weight of image definition and picture size is respectively set, and by the image of complete road image
Weight of the weighted sum result of clarity and picture size as the complete road image.It can be from determining complete road image
The middle maximum complete road image of weight selection;In the case where the maximum complete road image of weight includes multiple, from
A complete road image is randomly selected in the maximum complete road image of multiple weights.It, can also be from determining in practical application
It determines that weight is greater than the complete road image of weight threshold in complete road image, the full track of weight threshold is greater than from weight
A complete road image is chosen in the image of road.Optionally, the weight of image definition is less than the weight of picture size, at this time may be used
Illustrate that the priority of picture size is higher than the priority of image definition, for example chooses image ruler preferably from complete road image
The very little complete road image greater than size threshold.
As a kind of possible embodiment, the content that can be provided according to Fig. 8, determines the road using accurate detection algorithm
Second class road element of image and the second class road element information:
Step 801, first kind road element is intercepted from the road image, and it is at least one corresponding to obtain the road image
Screenshot includes at least one road element in first kind road element in the screenshot.
When it is implemented, intercepting first kind road element from road image, the corresponding screenshot of the road image is obtained, is had
Body can obtain section where each road element from each road element intercepted in first kind road element in road image
Figure, in this case, the number of road element is identical as the number of screenshot in the first road element.It certainly, can also be by the first kind
Multiple road elements in road element are truncated in same screenshot, here without limitation.Optionally, screenshot is to be sized
Image block, for example being sized can be pixel number for m*n, m and n, and the corresponding image block of screenshot is greater than mileage chart
Any one corresponding image block of road element as in.
Step 802, it using each screenshot as the input of accurate detection algorithm, obtains accurate the corresponding of detection algorithm output and cuts
The road element and road element information of figure.
When it is implemented, accurately being examined using the corresponding each screenshot of road image as the input of accurate detection algorithm
The road element and road element information of each screenshot of method of determining and calculating output.
Step 803, for every screenshot, by position of the road element in the screenshot in the road element information of the screenshot
It sets, is mapped as position of the road element in the road image.
When it is implemented, the position of road element is position of the road element in screenshot in the road element information of screenshot
It sets, in this step, needs position of the road element in screenshot being mapped as road element road image belonging to the screenshot
In position, specific mapping mode can be with are as follows: after intercepting first kind road element in road image, records road image pair
The screenshot is mapped to this according to position of the screenshot in the road image by position of each screenshot answered in the road image
Road image;According to position of the road element in screenshot, which is mapped as to the position in the road image.Than
Such as: being position view of the road element in screenshot in screenshot as illustrated in fig. 9, be screenshot in road as shown in figure 9b
Position view in image is as is shown in fig. 9 c position view of the road element in screenshot in road image,
In, the origin of coordinate system is the vertex in the screenshot upper left corner in Fig. 9 a, and transverse and longitudinal coordinate axis is respectively to connect with the vertex in the screenshot upper left corner
Two sides connect;In Fig. 9 b and Fig. 9 c the origin of coordinate system be the road image upper left corner vertex, transverse and longitudinal coordinate axis be respectively with
Two sides of the vertex connection in the road image upper left corner, position of the road element in screenshot are the corresponding rectangle of road element
Position of four vertex of detection block in screenshot, is respectively as follows: (a1, b1), (a2, b1), (a1, b2), (a2, b2), according to cut
Position of the figure in road image by the position that four vertex of screenshot map in road image be respectively as follows: (c1, d1), (c2,
D1), (c1, d2), (c2, d2), then, and position of four vertex of road element in road image are as follows: (a1+c1, b1+d1),
(a2+c1,b1+d1)、(a1+c1,b2+d1)、(a2+c1,b2+d1)。
Step 804, using the road element of each screenshot as the second class road element of the road image, and, it will be each
The position of the confidence level and corresponding road element of the road element in the road element information of screenshot in the road image is opened to make
For the second class road element information.
When it is implemented, using the road element of each screenshot as the second class road element of corresponding road image, it will
The confidence level of road element and position of the road element in road image after mapping in the road element information of each screenshot
It sets as the second class road element information.
First kind road element in road image is carried out screenshot, and cut what is obtained by this possible embodiment
Scheme the input as accurate detection algorithm, on the one hand ensure that comprising road element in every screenshot, to avoid the accurate inspection of input
Not the case where not including road element in the screenshot of method of determining and calculating, so that the processing for reducing accurate detection algorithm to a certain extent is negative
Load, on the other hand using screenshot as the input of accurate detection algorithm, reduces the size for inputting the image of accurate detection algorithm, from
And the processing speed that accurate detection algorithm determines road element information is improved to a certain extent.
As a kind of possible embodiment, the content provided according to Figure 10, intercepts described first from the road image
Class road element:
Step 1001, the first minimum circumscribed rectangle comprising all road elements in first kind road element is determined.
When it is implemented, being determined to all comprising this simultaneously for road element all in first kind road element
The minimum circumscribed rectangle of road element, as the first minimum circumscribed rectangle.
Step 1002, the length of the first minimum circumscribed rectangle is compared with length threshold, and, by the first minimum external square
The width of shape is compared with width threshold value.
When it is implemented, not being defined to the size of length threshold and width threshold value, length threshold can be less than width
Threshold value, optionally, length threshold are not less than width threshold value, and the size of length threshold and width threshold value can be according to practical application scene
Setting, here without limitation.
Step 1003, if the first minimum circumscribed rectangle is failed to grow up in the width of length threshold or the first minimum circumscribed rectangle
It then include the screenshot of all road elements in first kind road element from interception in the road image no more than width threshold value, it should
The length of screenshot is not less than above-mentioned length threshold, and the width of the screenshot is not less than above-mentioned width threshold value.
When it is implemented, the length in the first minimum circumscribed rectangle is less than or equal to length threshold, alternatively, first is minimum external
It include each road in first kind road element from interception in road image in the case that the frame of rectangle is less than or equal to width threshold value
The screenshot of road element.The length of the screenshot be greater than or equal to length threshold, also, the screenshot be wider than or be equal to width threshold value.
Step 1004, if the length of the first minimum circumscribed rectangle is greater than the roomy of length threshold or the first minimum circumscribed rectangle
In width threshold value, then the road element in first kind road element is grouped, obtains multiple groupings comprising each
The length of the second minimum circumscribed rectangle of all road elements is equal to length threshold, also, the second minimum circumscribed rectangle in grouping
Width is equal to width threshold value.
When it is implemented, the length in the first minimum circumscribed rectangle is greater than length threshold, alternatively, the first minimum circumscribed rectangle
In the case where being wider than width threshold value, the road element in first kind road element is grouped.For each of obtaining point
Group, comprising at least one road element in first kind road element in the grouping, also, include in the grouping all roads want
The length of second minimum circumscribed rectangle of element is equal to length threshold, also, the width of the second minimum circumscribed rectangle is equal to width threshold value.
It should be noted that in the embodiment of the present invention, the second minimum circumscribed rectangle be it is one long be equal to length threshold and
Width is equal to the rectangle of width threshold value.
It step 1005, include section of all road elements in the grouping from interception in the road image for each grouping
Figure.
When it is implemented, intercepting the corresponding screenshot of each grouping from road image, only wrapped in the screenshot of this step interception
Include all road elements in a grouping, that is, the screenshot quantity intercepted in this step from road image is identical as grouping number.
The length of screenshot is greater than or equal to length threshold in this step, and screenshot is wider than or is equal to width threshold value.
In this step, the corresponding screenshot of grouping, if there are same road element in the corresponding screenshot of different grouping,
Then determine the screenshot only comprising the part same road element, and by a part of same road element of this in the screenshot of the determination
Color is set to pure color, for example, being set to white.It wherein, will not completely include same road element, such as screenshot in different screenshots
It include road element A in 1, screenshot 2 also includes road element A, then, only have a screenshot in screenshot 1 or screenshot 2 and wraps completely
Road element A is included, another screenshot only includes part road element A, that is, is only able to display a part of road in another of road screenshot
Element A.
Optionally, the length of screenshot involved in the embodiment of the present application is equal to the product of length threshold and length variation, cuts
The width of figure is equal to the product of width threshold value and width difference.It can be inclined with length with length threshold centered on the center of rectangle
The product of difference intercepts screenshot, wherein right using the product of width threshold value and width difference as broadside for long side from road image
For step 1003, which is specially the first minimum circumscribed rectangle, and for step 1004, which is specially second
Minimum circumscribed rectangle.
In practical application, length variation and width difference can be determined in the following way:
Mode one: the sample road element for several road images that fast algorithm of detecting determines is advanced with;Statistics is each
The length and width of the corresponding theoretical hough transform frame of a sample road element, and, it is corresponding true to count each sample road element
The length and width of real hough transform frame;Determine the length of the corresponding theoretical hough transform frame of each sample road element and value, as
First and value, and, determine the length of the corresponding true rectangular detection block of each sample road element and value, as the second He
Value;By first and value with second and be worth difference absolute value, using the setting multiple of the absolute value as length variation;It determines each
The wide and value of the corresponding theoretical hough transform frame of a sample road element, as third and value, and, determine each sample road
The wide and value of the corresponding true rectangular detection block of road element, as the 4th and value;By third and value and the 4th and the difference of value
Absolute value, using the setting multiple of the absolute value as width difference.Wherein, setting multiple can be set according to practical application scene
It is fixed, for example multiple is set as 2, alternatively, setting multiple as 3.
Mode two: the sample road element for several road images that fast algorithm of detecting determines is advanced with;Statistics is each
The length and width of the corresponding theoretical hough transform frame of a sample road element, and, it is corresponding true to count each sample road element
The length and width of real hough transform frame;For each sample road element, the theoretical hough transform frame of the sample road element is determined
Length and true rectangular detection block length difference, as the first difference;The absolute value for taking first difference, as the sample road
The error in length of road element, and, determine the width and true rectangular detection block of the theoretical hough transform frame of the sample road element
Wide difference, as the second difference;The absolute value for taking second difference, the width error as the sample road element;It will
In the error in length of each sample road element, maximum error in length as length variation, and, each sample road element
Width error in, maximum width error is as width difference.
It should be noted that the length of boundary rectangle is equal to the width of boundary rectangle in the case where boundary rectangle is square,
The length of boundary rectangle and width are only used for distinguishing the different sides of rectangle in the embodiment of the present application, and the length of boundary rectangle is likely less than external
The width of rectangle, the length of boundary rectangle may also be greater than or equal to the width of boundary rectangle;The length and width of screenshot in the embodiment of the present application
It is only used for distinguishing the different sides of screenshot, the length of screenshot is likely less than the width of screenshot, and the length of screenshot may also be greater than or equal to screenshot
Width, wherein in the case where screenshot is square, the length and width of screenshot are equal, optionally, the side of screenshot and road image
Side is parallel.
As a kind of possible real-time mode, the road in first kind road element can be wanted according to the content that Figure 11 is provided
Element is grouped:
Step 1101, judge in first kind road element whether to include ungrouped road element, if so, executing step
1102, otherwise, execute step 1107.
Step 1102, the road element in ungrouped road element at road image top-left position is determined, as
Reference path element.
When it is implemented, being determined in road image most for still ungrouped road element in first kind road element
Left side and the road element being located at road image uppermost position, using the determining road element as in the road image
Road element at top-left position.
Step 1103, by the upper left corner of the corresponding hough transform frame of the reference path element and the second minimum circumscribed rectangle
Upper left angle alignment.
Step 1104, it if the second minimum circumscribed rectangle completely includes the corresponding hough transform frame of the reference path element, looks into
It looks in the second minimum circumscribed rectangle and whether completely includes the corresponding hough transform frame of other ungrouped road elements, if so, holding
Otherwise row step 1105 executes step 1106.
When it is implemented, can be according to the position of road element in the picture from left to right, sequence from top to bottom, to lookup
Whether other ungrouped road element is completely included in second minimum circumscribed rectangle.
Step 1105, the road element completely included in the second minimum circumscribed rectangle is divided to same grouping, and returned
Step 1101.
When it is implemented, the road element in same grouping can be stamped by identical mark, the road in different grouping is wanted
Element stamps different marks, can distinguish grouping belonging to road element in this way.
Step 1106, which forms a grouping, and return step 1101.
Step 1107, terminate the grouping to the road element in first kind road element.
Certainly, the content that Figure 11 is provided is only a kind of possible embodiment, also can use cluster in practical application and calculates
Method is grouped the road element in first kind road element, as long as guaranteeing the road element in obtained each grouping
To be completely contained in the second minimum circumscribed rectangle, also, all road elements have been grouped in first kind road element.
It is illustrated below with reference to device of the Figure 12 to determining road element information provided by the embodiments of the present application.
It as shown in figure 12, is the apparatus structure schematic diagram of determining road element information provided by the embodiments of the present application, comprising:
Module 1201 is obtained, for being directed to every determining road image, obtains and is somebody's turn to do using what fast algorithm of detecting determined
The first kind road element and first kind road element information of road image, and, acquisition is determined using accurate detection algorithm
Second class road element of the road image and the second class road element information, wherein road element information includes road element
Position in road image of confidence level and road element;
First determining module 1202, for determining phase between the first kind road element and the second class road element
Same road element;
Second determining module 1203, for being directed to each identical road element, if the first kind road element information
In the confidence level of the identical road element be greater than first threshold, also, this in the second class road element information is identical
The confidence level of road element is less than second threshold, then sets the identical road element in the first kind road element information
The position of the identical road element in the road image, is determined as this in reliability and the second class road element information
The road element information of identical road element, the first threshold are not less than the second threshold.
As a kind of possible embodiment, the device of determining road element information provided by the embodiments of the present application, is also wrapped
It includes:
Third determining module 1204, for determining road image in the following way:
According to the acquisition position of each candidate roads image, hierarchical cluster is carried out to the candidate roads image, obtains needle
To the first order classification results of candidate roads image, wherein the adjacent candidate roads figure of acquisition position in each first order classification
Actual range as between is not more than pre-determined distance;
For obtained each first order classification, however, it is determined that the adjacent candidate roads figure of acquisition time in first order classification
Road element as between is similar, then the adjacent candidate roads image of the acquisition time is divided to same classification, obtains second
Grade classification results;
A candidate roads image is chosen from the classification of each second level respectively, as road image.
As a kind of possible embodiment, the device of determining road element information provided by the embodiments of the present application, is also wrapped
It includes:
4th determining module 1205, for will collect before the third determining module 1204 determines road image
Input of each original image as the fast algorithm of detecting;Determine wanting comprising road for the fast algorithm of detecting output
The original image of element is candidate roads image.
As a kind of possible embodiment, the third determining module 1204, is specifically used for:
For the adjacent candidate roads image of acquisition time, the of the forward candidate roads image of acquisition time is obtained respectively
The corresponding first hough transform frame of one road element, and, the second road in the candidate roads image of acquisition time rearward is wanted
The corresponding second hough transform frame of element;
If the deviation of the Aspect Ratio of the first hough transform frame and the second hough transform frame preset range it
Interior, the described first hough transform frame is less than the second hough transform frame and the corresponding image district of the first hough transform frame
The characteristics of image of the characteristics of image in domain image-region corresponding with the second hough transform frame is similar, it is determined that described first
Road element is similar to the second road element.
As a kind of possible embodiment, the third determining module 1204 is specifically used in the following way from any
A candidate roads image is chosen in the classification of the second level:
For every candidate roads image in the classification of any second level, if each road element of the candidate roads image
Corresponding hough transform frame is completely contained in the candidate roads image, it is determined that the candidate roads image is complete mileage chart
Picture;
According to image definition and picture size, a complete road image is chosen from determining complete road image.
As a kind of possible embodiment, the third determining module 1204 is specifically used in the following way from determination
Complete road image in choose a complete road image:
For determining every complete road image, the image definition and picture size of the complete road image are determined
Weighted sum is as a result, weight as the complete road image;
The maximum complete road image of weight selection from determining complete road image.
As a kind of possible embodiment, the weight of described image clarity is less than the weight of described image size.
As a kind of possible embodiment, first determining module 1202, is specifically used for:
For each road element in first kind road element, determines in the second class road element and exist with the road element
Position in the road image has the road element of overlapping region, as coincidence road element;
If the road element be overlapped road element overlapping region be greater than region threshold, it is determined that the road element with should
Coincidence road element is identical road element.
As a kind of possible embodiment, the device of determining road element information provided by the embodiments of the present application, is also wrapped
It includes:
5th determining module 1206 determines the road image using accurate detection algorithm for realizing in the following way
Second class road element and the second class road element information:
The first kind road element is intercepted from the road image, is obtained the road image corresponding at least one and is cut
Scheme, includes at least one road element in first kind road element in the screenshot;
Using each Zhang Suoshu screenshot as the input of the accurate detection algorithm, the phase of the accurate detection algorithm output is obtained
Answer the road element and road element information of screenshot;
Position of the road element in the screenshot in the road element information of the screenshot is mapped as every screenshot
Position of the road element in the road image;
Using the road element of each screenshot as the second class road element of the road image, and, by each screenshot
The position of the confidence level of road element in road element information and corresponding road element in the road image is as the second class
Road element information.
As a kind of possible embodiment, the 5th determining module 1206 is specifically used in the following way from the road
The first kind road element is intercepted in the image of road:
Determine the first minimum circumscribed rectangle comprising all road elements in the first kind road element;
If first minimum circumscribed rectangle is failed to grow up in the width of length threshold or first minimum circumscribed rectangle
It then include section of all road elements in the first kind road element from interception in the road image no more than width threshold value
Figure, the length of the screenshot is not less than the length threshold, also, the width of the screenshot is not less than the width threshold value.
As a kind of possible embodiment, the 5th determining module 1206 is also used to:
If the length of first minimum circumscribed rectangle is greater than the length threshold or first minimum circumscribed rectangle
It is wider than the width threshold value, then the road element in first kind road element is grouped, obtains multiple groupings, wherein
The length of the second minimum circumscribed rectangle comprising road elements all in each grouping is equal to the length threshold, also, described the
The width of two small boundary rectangles is equal to the width threshold value;
It include the screenshot of all road elements in the grouping from interception in the road image for each grouping.
After the method and apparatus for describing determining road element information provided by the embodiments of the present application, this Shen is described below
Please embodiment provide nonvolatile computer storage media.
The embodiment of the present application provides a kind of nonvolatile computer storage media, and the computer storage medium is stored with can
Program is executed, which executes the side for realizing any determination road element information provided by the above embodiment
The step of method.
After the method, apparatus and the medium that describe determining road element information provided by the embodiments of the present application, it is situated between below
Continue computer equipment provided by the embodiments of the present application.
The embodiment of the present application also provides a kind of computer equipment, on a memory including memory, processor and storage
Computer program, the processor realize the side of any determination road element information in above-described embodiment when executing described program
The step of method.
The embodiment of the present application also provides a kind of computer equipment, for executing the letter of the determination road element in above-described embodiment
The method of breath, as shown in figure 13, for the hardware structural diagram of computer equipment described in the application implementation, which is set
It is standby to be specifically as follows desktop computer, portable computer, smart phone, tablet computer etc..Specifically, which can
To include memory 1301, the computer program of processor 1302 and storage on a memory, the processor execution journey
The step of method of any determination road element information in above-described embodiment is realized when sequence.Wherein, memory 1301 can wrap
Read-only memory (ROM) and random access memory (RAM) are included, and the journey stored in memory 1301 is provided to processor 1302
Sequence instruction and data.
Further, computer equipment described in the embodiment of the present application can also include input unit 1303 and output
Device 1304 etc..Input unit 1303 may include keyboard, mouse, touch screen etc.;Output device 1304 may include that display is set
It is standby, such as liquid crystal display (Liquid Crystal Display, LCD), cathode-ray tube (Cathode Ray Tube, CRT),
Touch screen etc..Memory 1301, processor 1302, input unit 1303 and output device 1304 can by bus or other
Mode connects, in Figure 13 for being connected by bus.
Processor 1302 calls the program instruction of the storage of memory 1301 and executes above-mentioned reality according to the program instruction of acquisition
The method that the determination road element information of example offer is provided.
Using method, apparatus, medium and the equipment of determining road element information provided by the embodiments of the present application, have following
The utility model has the advantages that
Fast algorithm of detecting and accurate detection algorithm belong to deep learning multi-target detection algorithm, consider to utilize quickly inspection
The relatively reliable and road element using the determination of accurate detection algorithm of the confidence level of the road element of method of determining and calculating determination is in the picture
Positional accuracy high these two aspects the advantages of, in the road element determined using fast algorithm of detecting and utilize accurate detection
When the road element that algorithm is determined is same road element, the same road element that will be determined using fast algorithm of detecting
Confidence level and using the position of road element in the picture determined using accurate detection algorithm as the same road
The road element information of element, to obtain accurate road element information.
It should be noted that although being referred to several modules of the device of determining road element information in the above detailed description,
But it is this division be only exemplary it is not enforceable.In fact, according to presently filed embodiment, it is above-described
The feature and function of two or more modules can embody in a module.Conversely, the spy of an above-described module
Function of seeking peace can be to be embodied by multiple modules with further division.
The application is referring to method, the process of equipment (system) and computer program product according to the embodiment of the present application
Figure and/or block diagram describe.It should be understood that every one stream in flowchart and/or the block diagram can be realized by computer program instructions
The combination of process and/or box in journey and/or box and flowchart and/or the block diagram.It can provide these computer programs
Instruct the processor of general purpose computer, special purpose computer, Embedded Processor or other programmable data processing devices to produce
A raw machine, so that being generated by the instruction that computer or the processor of other programmable data processing devices execute for real
The device for the function of being specified in present one or more flows of the flowchart and/or one or more blocks of the block diagram.
These computer program instructions, which may also be stored in, is able to guide computer or other programmable data processing devices with spy
Determine in the computer-readable memory that mode works, so that it includes referring to that instruction stored in the computer readable memory, which generates,
Enable the manufacture of device, the command device realize in one box of one or more flows of the flowchart and/or block diagram or
The function of being specified in multiple boxes.
These computer program instructions also can be loaded onto a computer or other programmable data processing device, so that counting
Series of operation steps are executed on calculation machine or other programmable devices to generate computer implemented processing, thus in computer or
The instruction executed on other programmable devices is provided for realizing in one or more flows of the flowchart and/or block diagram one
The step of function of being specified in a box or multiple boxes.
Although the preferred embodiment of the application has been described, it is created once a person skilled in the art knows basic
Property concept, then additional changes and modifications may be made to these embodiments.So it includes excellent that the following claims are intended to be interpreted as
It selects embodiment and falls into all change and modification of the application range.
Obviously, those skilled in the art can carry out various modification and variations without departing from the essence of the application to the application
Mind and range.In this way, if these modifications and variations of the application belong to the range of the claim of this application and its equivalent technologies
Within, then the application is also intended to include these modifications and variations.
Claims (14)
1. a kind of method of determining road element information characterized by comprising
For every determining road image, the first kind road for obtaining the road image determined using fast algorithm of detecting is wanted
Element and first kind road element information, and, obtain the second class road of the road image determined using accurate detection algorithm
Element and the second class road element information, wherein road element information includes that the confidence level of road element and road element exist
Position in road image;
Determine identical road element between the first kind road element and the second class road element;
For each identical road element, if the confidence level of the identical road element of this in the first kind road element information
Greater than first threshold, also, in the second class road element information the identical road element confidence level less than the second threshold
Value, then by the confidence level of the identical road element of this in the first kind road element information and the second class road element
Position of the identical road element of this in information in the road image is determined as the road element letter of the identical road element
Breath, the first threshold are not less than the second threshold.
2. the method according to claim 1, wherein determining road image in the following way:
According to the acquisition position of each candidate roads image, hierarchical cluster is carried out to the candidate roads image, is obtained for time
Select the first order classification results of road image, wherein in the classification of each first order the adjacent candidate roads image of acquisition position it
Between actual range be not more than pre-determined distance;
For obtained each first order classification, however, it is determined that in first order classification the adjacent candidate roads image of acquisition time it
Between road element it is similar, then the adjacent candidate roads image of the acquisition time is divided to same classification, obtains the second fraction
Class result;
A candidate roads image is chosen from the classification of each second level respectively, as road image.
3. according to the method described in claim 2, it is characterized in that, before determining road image, further includes:
Using collected each original image as the input of the fast algorithm of detecting;
The original image comprising road element for determining the fast algorithm of detecting output is candidate roads image.
4. according to the method described in claim 2, it is characterized in that, determining the candidate that acquisition time is adjacent in first order classification
Road element between road image is similar, specifically includes:
For the adjacent candidate roads image of acquisition time, first of the forward candidate roads image of acquisition time is obtained respectively
The corresponding first hough transform frame of road element, and, the second road element pair in the candidate roads image of acquisition time rearward
The the second hough transform frame answered;
If the deviation of the Aspect Ratio of the first hough transform frame and the second hough transform frame is within preset range, institute
The first hough transform frame is stated less than the second hough transform frame and the corresponding image-region of the first hough transform frame
The characteristics of image of characteristics of image image-region corresponding with the second hough transform frame is similar, it is determined that first road is wanted
It is plain similar to the second road element.
5. according to the method described in claim 2, it is characterized in that, choosing a candidate roads figure from the classification of any second level
Picture specifically includes:
For every candidate roads image in any second level classification, if each road element of the candidate roads image
Corresponding hough transform frame is completely contained in the candidate roads image, it is determined that the candidate roads image is complete mileage chart
Picture;
According to image definition and picture size, a complete road image is chosen from determining complete road image.
6. according to the method described in claim 5, it is characterized in that, choosing a full track from determining complete road image
Road image, specifically includes:
For determining every complete road image, the weighting of the image definition and picture size of the complete road image is determined
Summed result, the weight as the complete road image;
The maximum complete road image of weight selection from determining complete road image.
7. according to the method described in claim 6, it is characterized in that, the weight of described image clarity is less than described image size
Weight.
8. the method according to claim 1, wherein determining the first kind road element and second class road
Identical road element between the element of road, specifically includes:
For each road element in first kind road element, determine in the second class road element with the road element in the road
Position in the image of road has the road element of overlapping region, as coincidence road element;
If the road element is greater than region threshold with the overlapping region for being overlapped road element, it is determined that the road element is overlapped with this
Road element is identical road element.
9. the method according to claim 1, wherein determining the second of the road image using accurate detection algorithm
Class road element and the second class road element information, specifically include:
The first kind road element is intercepted from the road image, obtains the corresponding at least screenshot of the road image, institute
It states in screenshot including at least one road element in first kind road element;
Using each Zhang Suoshu screenshot as the input of the accurate detection algorithm, obtains accurate the corresponding of detection algorithm output and cut
The road element and road element information of figure;
Position of the road element in the screenshot in the road element information of the screenshot is mapped as road for every screenshot
Position of the element in the road image;
Using the road element of each screenshot as the second class road element of the road image, and, by the road of each screenshot
The position of the confidence level of road element in element information and corresponding road element in the road image is as the second class road
Element information.
10. according to the method described in claim 9, being wanted it is characterized in that, intercepting the first kind road from the road image
Element specifically includes:
Determine the first minimum circumscribed rectangle comprising all road elements in the first kind road element;
If first minimum circumscribed rectangle fail to grow up in length threshold or first minimum circumscribed rectangle width less
It then include the screenshot of all road elements in the first kind road element, institute from interception in the road image in width threshold value
The length of screenshot is stated not less than the length threshold, also, the width of the screenshot is not less than the width threshold value.
11. according to the method described in claim 10, it is characterized by further comprising:
If the length of first minimum circumscribed rectangle is greater than the roomy of the length threshold or first minimum circumscribed rectangle
In the width threshold value, then the road element in first kind road element is grouped, obtain it is multiple grouping comprising
The length of the second minimum circumscribed rectangle of all road elements is equal to the length threshold in each grouping, also, described second small
The width of boundary rectangle is equal to the width threshold value;
It include the screenshot of all road elements in the grouping from interception in the road image for each grouping.
12. a kind of device of determining road element information characterized by comprising
Module is obtained, for obtaining the road image determined using fast algorithm of detecting for every determining road image
First kind road element and first kind road element information, and, obtain using accurate detection algorithm determine the mileage chart
Second class road element of picture and the second class road element information, wherein road element information includes the confidence level of road element
And position of the road element in road image;
First determining module, for determining identical road between the first kind road element and the second class road element
Element;
Second determining module, for being directed to each identical road element, if this in the first kind road element information is identical
The confidence level of road element be greater than first threshold, also, identical road element in the second class road element information
Confidence level be less than second threshold, then by the confidence level of the identical road element of this in the first kind road element information and
The position of the identical road element in the road image, is determined as the identical road in the second class road element information
The road element information of road element, the first threshold are not less than the second threshold.
13. a kind of nonvolatile computer storage media, which is characterized in that the computer storage medium is stored with executable journey
Sequence, the executable code processor execute the step of realizing claim 1-11 any the method.
14. a kind of computer equipment, which is characterized in that including memory, the computer journey of processor and storage on a memory
The step of sequence, the processor realizes claim 1-11 any the method when executing described program.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810322044.4A CN110377670B (en) | 2018-04-11 | 2018-04-11 | Method, device, medium and equipment for determining road element information |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810322044.4A CN110377670B (en) | 2018-04-11 | 2018-04-11 | Method, device, medium and equipment for determining road element information |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110377670A true CN110377670A (en) | 2019-10-25 |
CN110377670B CN110377670B (en) | 2021-11-26 |
Family
ID=68242957
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810322044.4A Active CN110377670B (en) | 2018-04-11 | 2018-04-11 | Method, device, medium and equipment for determining road element information |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110377670B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111582013A (en) * | 2019-12-27 | 2020-08-25 | 珠海大横琴科技发展有限公司 | Ship retrieval method and device based on gray level co-occurrence matrix characteristics |
CN112149624A (en) * | 2020-10-16 | 2020-12-29 | 腾讯科技(深圳)有限公司 | Traffic identification image processing method and device |
CN112595728A (en) * | 2021-03-03 | 2021-04-02 | 腾讯科技(深圳)有限公司 | Road problem determination method and related device |
CN113936458A (en) * | 2021-10-12 | 2022-01-14 | 中国联合网络通信集团有限公司 | Method, device, equipment and medium for judging congestion of expressway |
Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2000132093A (en) * | 1998-08-19 | 2000-05-12 | Ec Service:Kk | Road information management system and its method |
US20130054106A1 (en) * | 2011-08-22 | 2013-02-28 | Honda Research Institute Europe Gmbh | Method and system for predicting movement behavior of a target traffic object |
CN103366156A (en) * | 2012-04-09 | 2013-10-23 | 通用汽车环球科技运作有限责任公司 | Road structure detection and tracking |
CN104361142A (en) * | 2014-12-12 | 2015-02-18 | 华北水利水电大学 | Detection method for rapid change in multi-source navigation electronic map vector road network |
US20150332097A1 (en) * | 2014-05-15 | 2015-11-19 | Xerox Corporation | Short-time stopping detection from red light camera videos |
US20160063858A1 (en) * | 2014-08-29 | 2016-03-03 | Honda Research Institute Europe Gmbh | Method and system for using global scene context for adaptive prediction and corresponding program, and vehicle equipped with such system |
CN106650641A (en) * | 2016-12-05 | 2017-05-10 | 北京文安智能技术股份有限公司 | Traffic light positioning and identification method, device and system |
US20170193384A1 (en) * | 2016-01-06 | 2017-07-06 | GM Global Technology Operations LLC | Determining driver intention at traffic intersections for automotive crash avoidance |
CN107423757A (en) * | 2017-07-14 | 2017-12-01 | 北京小米移动软件有限公司 | clustering processing method and device |
CN107463918A (en) * | 2017-08-17 | 2017-12-12 | 武汉大学 | Lane line extracting method based on laser point cloud and image data fusion |
CN107690659A (en) * | 2016-12-27 | 2018-02-13 | 深圳前海达闼云端智能科技有限公司 | A kind of image identification system and image-recognizing method |
US20180181817A1 (en) * | 2015-09-10 | 2018-06-28 | Baidu Online Network Technology (Beijing) Co., Ltd. | Vehicular lane line data processing method, apparatus, storage medium, and device |
-
2018
- 2018-04-11 CN CN201810322044.4A patent/CN110377670B/en active Active
Patent Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2000132093A (en) * | 1998-08-19 | 2000-05-12 | Ec Service:Kk | Road information management system and its method |
US20130054106A1 (en) * | 2011-08-22 | 2013-02-28 | Honda Research Institute Europe Gmbh | Method and system for predicting movement behavior of a target traffic object |
CN103366156A (en) * | 2012-04-09 | 2013-10-23 | 通用汽车环球科技运作有限责任公司 | Road structure detection and tracking |
US20150332097A1 (en) * | 2014-05-15 | 2015-11-19 | Xerox Corporation | Short-time stopping detection from red light camera videos |
US20160063858A1 (en) * | 2014-08-29 | 2016-03-03 | Honda Research Institute Europe Gmbh | Method and system for using global scene context for adaptive prediction and corresponding program, and vehicle equipped with such system |
CN104361142A (en) * | 2014-12-12 | 2015-02-18 | 华北水利水电大学 | Detection method for rapid change in multi-source navigation electronic map vector road network |
US20180181817A1 (en) * | 2015-09-10 | 2018-06-28 | Baidu Online Network Technology (Beijing) Co., Ltd. | Vehicular lane line data processing method, apparatus, storage medium, and device |
US20170193384A1 (en) * | 2016-01-06 | 2017-07-06 | GM Global Technology Operations LLC | Determining driver intention at traffic intersections for automotive crash avoidance |
CN106650641A (en) * | 2016-12-05 | 2017-05-10 | 北京文安智能技术股份有限公司 | Traffic light positioning and identification method, device and system |
CN107690659A (en) * | 2016-12-27 | 2018-02-13 | 深圳前海达闼云端智能科技有限公司 | A kind of image identification system and image-recognizing method |
CN107423757A (en) * | 2017-07-14 | 2017-12-01 | 北京小米移动软件有限公司 | clustering processing method and device |
CN107463918A (en) * | 2017-08-17 | 2017-12-12 | 武汉大学 | Lane line extracting method based on laser point cloud and image data fusion |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111582013A (en) * | 2019-12-27 | 2020-08-25 | 珠海大横琴科技发展有限公司 | Ship retrieval method and device based on gray level co-occurrence matrix characteristics |
CN112149624A (en) * | 2020-10-16 | 2020-12-29 | 腾讯科技(深圳)有限公司 | Traffic identification image processing method and device |
CN112595728A (en) * | 2021-03-03 | 2021-04-02 | 腾讯科技(深圳)有限公司 | Road problem determination method and related device |
CN113936458A (en) * | 2021-10-12 | 2022-01-14 | 中国联合网络通信集团有限公司 | Method, device, equipment and medium for judging congestion of expressway |
Also Published As
Publication number | Publication date |
---|---|
CN110377670B (en) | 2021-11-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110377670A (en) | A kind of method, apparatus, medium and the equipment of determining road element information | |
US9349189B2 (en) | Occlusion resistant image template matching using distance transform | |
CN105122308B (en) | System and method for using the multichannel biological marker of the structural unicellular division of continuous dyeing quantitative | |
CN112084869B (en) | Compact quadrilateral representation-based building target detection method | |
CN110490202A (en) | Detection model training method, device, computer equipment and storage medium | |
US20150347478A1 (en) | Systems and methods for context-aware and personalized access to visualizations of road events | |
CN105975929A (en) | Fast pedestrian detection method based on aggregated channel features | |
CN109685055A (en) | Text filed detection method and device in a kind of image | |
US9158992B2 (en) | Acceleration of linear classifiers | |
CN105608417A (en) | Traffic signal lamp detection method and device | |
CN105608454A (en) | Text structure part detection neural network based text detection method and system | |
CN115294473A (en) | Insulator fault identification method and system based on target detection and instance segmentation | |
US20220357176A1 (en) | Methods and data processing systems for predicting road attributes | |
CN106778633B (en) | Pedestrian identification method based on region segmentation | |
CN107710280A (en) | Object method for visualizing | |
CN110533039A (en) | A kind of true-false detection method of license plate, device and equipment | |
CN111860494A (en) | Optimization method and device for image target detection, electronic equipment and storage medium | |
CN104182757A (en) | Method of acquiring actual coverage area of measured target and device | |
CN105809108B (en) | Pedestrian's localization method and system based on distributed vision | |
CN105405130A (en) | Cluster-based license image highlight detection method and device | |
CN111324616A (en) | Method, device and equipment for detecting lane line change information | |
CN114998744A (en) | Agricultural machinery track field segmentation method based on motion and vision dual-feature fusion | |
CN110738076A (en) | People counting method and system in images | |
Ghani et al. | Accuracy Assessment of Urban Growth Pattern Classification Methods Using Confusion Matrix and ROC Analysis | |
CN107610224A (en) | It is a kind of that algorithm is represented based on the Weakly supervised 3D automotive subjects class with clear and definite occlusion modeling |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |