CN109583329A - Winding detection method based on the screening of road semanteme road sign - Google Patents
Winding detection method based on the screening of road semanteme road sign Download PDFInfo
- Publication number
- CN109583329A CN109583329A CN201811350086.5A CN201811350086A CN109583329A CN 109583329 A CN109583329 A CN 109583329A CN 201811350086 A CN201811350086 A CN 201811350086A CN 109583329 A CN109583329 A CN 109583329A
- Authority
- CN
- China
- Prior art keywords
- road sign
- image
- road
- original
- similar
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
- G06V20/582—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of traffic signs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention relates to a kind of winding detection methods based on the screening of semantic road sign, extract a certain number of original road signs in image, and as the description to picture, the original road sign includes dynamic guidepost and static road sign；Screen the dynamic guidepost that semantic information is had in original road sign；Feature extraction is carried out to the original road sign after screening；It treats query image and image to be retrieved carries out the above operation, determine similar road sign of each road sign in image to be retrieved in image to be checked；Calculate image to be checked and similar road sign similarity score between images, winding is determined according to similarity score.The present invention has obtained static road sign, has reduced description of the multidate information as road sign to picture in the technology for increasing the screening of road semanteme road sign, this makes final recognition result also more accurate, more robust.
Description
Technical field
The invention belongs to technical field of computer vision, the detection more particularly to one kind for being related to road semanteme road sign are based on
The winding detection method of road semanteme road sign screening.
Background technique
In the independent navigation research of robot, winding detection has as instant positioning and the rear end of map structuring technology
Very important effect.Winding detection is main to be to solve the problem of robot pose drifts about at any time, and usual winding detects institute
The winding information of offer can correct constructed by robot there are error even deform track map, with auxiliary robot into
Row precision navigation.In recent years, the research of outdoor robot and unmanned vehicle navigation achieved quick development, outdoor complicated and changeable
Environment new higher requirement is proposed to the robustness of winding detection technique.
There are many outstanding algorithms in the research progress that winding detection road sign extracts, for example, using BTNG algorithm,
Edgebox algorithm, Selective search algorithm etc. extract road sign in picture.The overall situation that these methods not only take into account picture is special
Sign description and local feature description, and good effect is also shown in complex environment.But in the ring based on road
These methods still have the space of promotion in border, and common ground is all that the original road sign of fixed quantity is extracted in picture including dynamic
State road sign and static road sign.Since dynamic guidepost is variable, if dynamic guidepost participates in the similarity calculated between image
Scoring then generates error hiding to will lead to judgement to winding, at the same this method during road sign extracts due to dynamic road
It is low to there is will lead to road sign repeatability in target.
Summary of the invention
It is inaccurate, poor robustness it is an object of the present invention to solve the winding result that the above-mentioned prior art detects
Problem provides a kind of winding detection method based on the screening of road semanteme road sign.
The technical solution adopted by the present invention to solve the technical problems is: the winding detection based on the screening of road semanteme road sign
Method extracts a certain number of original road signs in image, as the description to picture；The original road sign include dynamic guidepost and
Static road sign；Screen the dynamic guidepost that semantic information is had in original road sign；Feature is carried out to original road sign remaining after screening
It extracts；It treats query image and image to be retrieved carries out the above operation, determine that each road sign is to be retrieved in image to be checked
Similar road sign in image；Calculate image to be checked and similar road sign similarity score between images, according to similarity
It scores and determines winding.
In the present invention, link is screened in dynamic guidepost, present invention uses the thoughts of juxtaposition area, that is, dynamic road
Whether mark is screened depending on the juxtaposition area between dynamic guidepost and original road sign.It is this to be based on calculating juxtaposition face
Long-pending method effectively reduces the error hiding as caused by multidate information, and multidate information is more in image, remaining after screening
Original road specificity will be stronger, and error hiding rate is also lower, therefore the accuracy rate of winding detection and robustness all obtain
It significantly improves, this instant technical support for positioning with map structuring providing the foundation property outdoor for robot.
Preferably, a certain number of original road signs in the extraction image, specifically include: extracting and specified on every image
The road sign of number remembers that specified number is N, the mark information that every figure extractsIt is expressed asIts
Middle r indicates that the road sign is located on the image of serial number r, and i reflects serial number of this road sign in image r,
WithThe coordinate in the road sign upper left corner is respectively indicated, it is wide and high；Assuming that shared T images, then the value of r arrives T integer, i for 1
The integer for arriving N for 1.
Preferably, the mark information of the dynamic guidepost with semantic informationIt is expressed asWherein r indicates that the dynamic guidepost is located on the image of serial number r, and J reflects this dynamic guidepost
Serial number in image r,WithThe coordinate in the dynamic guidepost upper left corner is respectively indicated, it is wide and high；
Assuming that shared T images, then the value of r is 1 integer for arriving T, it is assumed that the maximum quantity that road sign is set out in identification is R, then J takes
The integer that value arrives R for 1.
Preferably, whether the dynamic guidepost is screened depending on the juxtaposition between dynamic guidepost and original road sign
Area；If the ratio between the juxtaposition area and original road sign area is less than or equal to given threshold, the road sign
Reservation will be selected；If the ratio between the juxtaposition area and original road sign area is greater than given threshold, should
Road sign will be identified and screen out.
Preferably, the process of the screening dynamic guidepost specifically includes:
Calculate the juxtaposition area of original road sign and dynamic guidepost:
Overlap=width × height
Wherein, endx indicates the maximum value of original road sign Yu dynamic guidepost upper right corner X-coordinate, and startx indicates original road sign
With the minimum value of dynamic guidepost upper left corner X-coordinate, width indicates the width of original road sign Yu dynamic guidepost juxtaposition area,
Endy indicates the maximum value of original road sign Yu dynamic guidepost lower left corner Y-coordinate, and starty indicates that original road sign and dynamic guidepost are left
The minimum value of upper angle Y-coordinate, height indicate the height of original road sign Yu dynamic guidepost juxtaposition area, and overlap is indicated
The juxtaposition area of original road sign and dynamic guidepost, Area indicate that the area of original road sign, ratio indicate juxtaposition area
The shared specific gravity in original road sign；
If overlap is less than or equal to 0, illustrate there is no multidate information in original road sign, original road sign will be by this time
Retain；If overlap is greater than 0, at this time given threshold β, if ratio is greater than β, illustrate the multidate information in original road sign
It is affected to entire road sign, original road sign will be removed, if ratio is less than β, original road sign will be retained
It is specifically included preferably, carrying out feature extraction to original road sign remaining after screening:
Remaining original road sign after screening is extracted, the fixed number of fetching is F, and the size of F road sign is normalized；
Original road sign after normalization is input in convolutional Neural metanetwork model AlexNet, AlexNet middle layer is taken
Description as each road sign is exported, and its vector is turned to description of 9216 dimensions；
The dimension of description of 9216 dimension is reduced using gaussian random projecting method, the dimension after reduction is 1024,
Then it carries out again unitization, obtains the final feature vector of road sign, remember that the feature vector of road sign isThe value of t arrives F's for 1
Integer.
Preferably, similar road sign of each road sign in image to be retrieved refers in the determination image to be checked, choosing
Road sign similar to inquiry road sign in road sign to be retrieved is taken to screen as candidate similar road sign, then to candidate similar road sign,
Road sign similar to inquiry road sign feature and shape dissmilarity in the candidate similar road sign of removal, remaining road sign is as similar road sign.
Preferably, it is described calculate image to be checked and similar road sign similarity score between images, according to phase
Like degree score determine winding, in particular to:
First calculate the shape similarity S between road signig:
The value range of wherein i, g are 1 integer for arriving F, and the value range of a, b are 1 integer for arriving T；Then road sign is calculated
Between characteristic similarity dig:
The overall similarity S between picture is finally calculated againab:
Wherein Fa, FbRespectively indicate picture a, the road sign quantity in b；
It is focused to find out the maximum one group of matching (a, b) of scoring in candidate image, if image a to be retrieved and candidate image b
Similarity score is greater than the decision threshold λ of setting, then image a and candidate image b to be retrieved are shooting in the figure of the same scene
Picture, as a winding.
Preferably, the similar road sign refers to: (1) inquiry road sign feature being mapped to the Hash by hash function
In table, finds by the nearest neighbor search method based on COS distance in Hash table and inquire most like the one before of road sign feature
Road sign is as candidate similar road sign；(2) number for obtaining the similar road sign of the candidate, it is similar to find each candidate by inquiry table
The corresponding mark information of road sign；(3) by shape similarity inspection, road similar with inquiry road sign in candidate similar road sign is chosen
It is denoted as similar road sign.
Preferably, the shape similarity inspection refers to:
For a certain inquiry road sign c and the similar road sign z of a candidate, it is considered if the two size meets following relationship
It is similar:
Simultaneously
Wherein Wc, hcRespectively indicate the width and height of road sign c, Wz, WzRespectively indicate the width and height of road sign z.
Substantial effect of the invention: the present invention is in visual scene recognizer of the tradition based on convolutional Neural metanetwork
On the basis of increase road semanteme road sign screening technology, obtained static road sign, reduced multidate information as road sign to figure
The description of piece, this makes final recognition result also more accurate, more robust.
Detailed description of the invention
Fig. 1 is a kind of flow diagram of the invention.
Specific embodiment
Below by specific example, and in conjunction with attached drawing, technical scheme of the present invention will be further explained in detail.
Embodiment:
Based on the winding detection method of road semanteme road sign screening, a certain number of original road signs in image are extracted, as
Description to picture；The original road sign includes dynamic guidepost and static road sign；It screens and has semantic information in original road sign
Dynamic guidepost；Feature extraction is carried out to original road sign remaining after screening；Treat query image and image to be retrieved carry out with
Upper operation determines similar road sign of each road sign in image to be retrieved in image to be checked；Calculate image to be checked to it is similar
Road sign similarity score between images, winding is determined according to similarity score.As shown in Figure 1, early period is examined using target
Method of determining and calculating Edgeboxs extracts the original road sign of specified quantity on image to be checked and image to be retrieved, and uses algorithm simultaneously
Faster rcnn extracts dynamic guidepost on the image, and then being rejected using the thought of juxtaposition has the original of multidate information
Road sign, it is each that remaining dynamic guidepost is utilized to existing convolutional neural networks model AlexNet after centainly pre-processing
Road sign generates corresponding description, and in order to improve the computational efficiency of subsequent match, early period has also carried out dimensionality reduction, later period to description
Using the similar road sign corresponding to the road sign in hash algorithm retrieval and inquisition image in image to be retrieved, finally calculate image it
Between overall similarity, judge whether there is winding and provide the position of winding.Here image to be retrieved is quite schemed with one
As library, winding detection is briefly exactly that image identical with image to be checked is searched in image to be retrieved.Specifically include as
Lower step:
1) the original road sign specified number on every picture in data set is extracted using algorithm of target detection Edgeboxs, is referred to
Fixed number mesh is N, and N takes 300 in this example.The mark information that every figure extractsTable is notWherein r
Indicating that the road sign is located on the image of serial number r, i reflects serial number of this road sign in image r,WithThe coordinate in the road sign upper left corner is respectively indicated, it is wide and high.Assuming that shared T images, then the value of r is 1 to T integer, i is
1 arrives the integer of N.
2) it using the dynamic guidepost on every picture in algorithm Faster RCNN identification data set, identifies on every image
Dynamic guidepost (such as pedestrian, vehicle), every image recognition go out mark informationIt is expressed asWherein r indicates that the dynamic guidepost is located on the image of serial number r, and J reflects this dynamic guidepost
Serial number in image r, KindThe coordinate in the dynamic guidepost upper left corner is respectively indicated, it is wide and high, it is false
If shared T images, then the value of r is 1 integer for arriving T, it is assumed that the maximum value for identifying dynamic guidepost quantity is R, then J
Value is 1 integer for arriving R.
3) the original road sign and dynamic guidepost of every image are obtained from 1) 2), the thought of juxtaposition area is utilized to carry out
Screening.
Overlap=width × height
Using above-mentioned formula, overlap indicates the juxtaposition area of original road sign and dynamic guidepost.Area indicates original
The area of road sign, ratio indicate juxtaposition area specific gravity shared in original road sign.If overlap is less than or equal to
0, illustrate there is no multidate information in original road sign, original road sign will be retained at this time.If overlap is greater than 0, set at this time
Determine threshold value beta, β takes 0.1 in this example, if ratio is greater than β, illustrates the multidate information in original road sign to the shadow of entire road sign
Sound is larger, and original road sign will be removed, if ratio is less than β, original road sign will be retained.
4) 3) remaining original road sign in is taken and is specified number after F is 50 road signs normalization, be input to convolutional Neural net
In network model AlexNet, description of the output of AlexNet middle layer Pool5 as each road sign, Pool5 layers of output layer are taken
Vector is needed again to turn to description of 9216 dimensions.
5) the sub- dimensions of description of 4) middle 9216 dimensions extracted are reduced using gaussian random projecting method, is reduced to 1024
Dimension, and be denoted asThe integer that the value range of t is 1 to 50 is so that it is convenient to subsequent similarity calculation.
6) hash algorithm is utilized for each road sign in query image, determines their the similar roads in image to be retrieved
Mark.
Inquiry road sign feature is mapped in the Hash table by hash function, in Hash table by be based on cosine away from
From nearest neighbor search method find it is most like to inquiry road sign feature before several road signs as candidate similar road sign；It obtains
The number of the similar road sign of the candidate, finds the corresponding mark information of the similar road sign of each candidate by inquiry table；Pass through shape
Similarity inspection, removal feature is similar and the road sign of shape dissmilarity, choose in candidate similar road sign with inquire road sign feature phase
Seemingly and the similar road sign of shape is as similar road sign.
For a certain inquiry road sign c and the similar road sign z of a candidate, it is considered if the two size meets following relationship
It is similar:
Simultaneously
Wherein Wc, hcRespectively indicate the width and height of road sign c, Wz, WzRespectively indicate the width and height of road sign z.
7) similarity score between query image and image to be retrieved is calculated, highest a pair of of the image and being greater than of scoring refers to
Determine threshold value, can determine whether to be a winding.
The similarity of picture entirety, including calculate mark shape similarity S between pictureig, the similarity d of road sign featureig, tool
Body calculates as follows:
The integer that the value range of wherein a, b are 1 to 4000, Fa, FbRespectively indicate picture a, the road sign quantity in b, this reality
Apply F in examplea, FbAll take 50.
It is focused to find out the maximum one group of matching (a, b) of scoring in candidate image, if image a to be retrieved and candidate image b
Similarity score is greater than the decision threshold λ of setting, then image a and candidate image b to be retrieved are shooting in the figure of the same scene
Picture, as a winding.
Embodiment described above is a kind of preferable scheme of the invention, and the purpose for publicizing and implementing example is further to manage
The solution present invention.Or else under the premise of exceeding technical solution documented by claim, various substitutions and modifications are all possible.
Claims (10)
1. the winding detection method based on the screening of road semanteme road sign, which is characterized in that
A certain number of original road signs in image are extracted, as the description to picture；The original road sign include dynamic guidepost and
Static road sign；
Screen the dynamic guidepost that semantic information is had in original road sign；
Feature extraction is carried out to original road sign remaining after screening；
It treats query image and image to be retrieved carries out the above operation, determine that each road sign is in figure to be retrieved in image to be checked
Similar road sign as in；
Calculate image to be checked and similar road sign similarity score between images, winding is determined according to similarity score.
2. the winding detection method as described in claim 1 based on the screening of road semanteme road sign, which is characterized in that the extraction
A certain number of original road signs in image, specifically include:
The road sign specified number on every image is extracted, remembers that specified number is N, the mark information that every figure extractsIt indicates
ForWherein r indicates that the road sign is located on the image of serial number r, and i reflects this road sign in image r
In serial number,WithThe coordinate in the road sign upper left corner is respectively indicated, it is wide and high；Assuming that shared T images, that
The value of r is 1 to T integer, and i is 1 integer for arriving N.
3. the winding detection method as described in claim 1 based on the screening of road semanteme road sign, which is characterized in that described to have
The mark information of the dynamic guidepost of semantic informationIt is expressed asWherein r indicates the dynamic guidepost
On the image of serial number r, J reflects serial number of this dynamic guidepost in image r,WithPoint
The coordinate in the dynamic guidepost upper left corner is not indicated, it is wide and high；Assuming that shared T images, then the value of r is 1 integer for arriving T,
Assuming that the maximum quantity that road sign is set out in identification is R, then the value of J is 1 integer for arriving R.
4. the winding detection method as described in claim 1 based on the screening of road semanteme road sign, which is characterized in that the dynamic
Whether road sign is screened depending on the juxtaposition area between dynamic guidepost and original road sign；If the juxtaposition area
Ratio between original road sign area is less than or equal to given threshold, then the road sign will be selected reservation；If the friendship
The ratio pitched between overlapping area and original road sign area is greater than given threshold, then the road sign will be identified and screen out.
5. the winding detection method as claimed in claim 4 based on the screening of road semanteme road sign, which is characterized in that the screening
The process of dynamic guidepost specifically includes:
Calculate the juxtaposition area of original road sign and dynamic guidepost:
Overlap=width × height
Wherein, endx indicates the maximum value of original road sign Yu dynamic guidepost upper right corner X-coordinate, and startx indicates original road sign and moves
The minimum value of state road sign upper left corner X-coordinate, width indicate the width of original road sign Yu dynamic guidepost juxtaposition area, endy
Indicate the maximum value of original road sign Yu dynamic guidepost lower left corner Y-coordinate, starty indicates original road sign and dynamic guidepost upper left corner Y
The minimum value of coordinate, height indicate the height of original road sign Yu dynamic guidepost juxtaposition area, and overlap indicates original road
The juxtaposition area of mark and dynamic guidepost, Area indicate that the area of original road sign, ratio indicate juxtaposition area original
Shared specific gravity in road sign；
If overlap is less than or equal to 0, illustrate there is no multidate information in original road sign, original road sign will be retained at this time；
If overlap is greater than 0, at this time given threshold β, if ratio is greater than β, illustrate the multidate information in original road sign
It is affected to entire road sign, original road sign will be removed, if ratio is less than β, original road sign will be retained.
6. the winding detection method as described in claim 1 based on the screening of road semanteme road sign, which is characterized in that after screening
Remaining original road sign carries out feature extraction and specifically includes:
Remaining original road sign after screening is extracted, the fixed number of fetching is F, and the size of F road sign is normalized；
Original road sign after normalization is input in convolutional Neural metanetwork model AlexNet, AlexNet middle layer is taken to export
As description of each road sign, and the description that its vector is turned to 9216 dimensions is sub；
The dimension of description of 9216 dimension is reduced using gaussian random projecting method, the dimension after reduction is 1024, then
It carries out again unitization, obtains the final feature vector of road sign, remember that the feature vector of road sign isThe value of t arrives the whole of F for 1
Number.
7. the winding detection method as described in claim 1 based on the screening of road semanteme road sign, which is characterized in that the determination
Similar road sign of each road sign in image to be retrieved refers in image to be checked, choose in road sign to be retrieved with inquiry road sign phase
As road sign screened as candidate similar road sign, then to candidate similar road sign, in the candidate similar road sign of removal with inquiry road
Mark that feature is similar and the road sign of shape dissmilarity, remaining road sign is as similar road sign.
8. the winding detection method as described in claim 1 based on the screening of road semanteme road sign, which is characterized in that the calculating
Image to be checked and similar road sign similarity score between images, according to similarity score determine winding, in particular to:
First calculate the shape similarity S between road signig:
The value range of wherein i, g are 1 integer for arriving F, and the value range of a, b are 1 integer for arriving T；
Then the characteristic similarity d between road sign is calculatedig:
The overall similarity S between picture is finally calculated againab:
Wherein Fa, FbRespectively indicate picture a, the road sign quantity in b；
It is focused to find out the maximum one group of matching (a, b) of scoring in candidate image, if image a's to be retrieved and candidate image b is similar
Degree scoring is greater than the decision threshold λ of setting, then image a and candidate image b to be retrieved are shooting in the image of the same scene, i.e.,
For a winding.
9. the winding detection method as claimed in claim 7 based on the screening of road semanteme road sign, which is characterized in that described similar
The acquisition methods of road sign specifically include:
(1) inquiry road sign feature is mapped in the Hash table by hash function, in Hash table by be based on cosine away from
From nearest neighbor search method find it is most like to inquiry road sign feature before several road signs as candidate similar road sign；
(2) number for obtaining the similar road sign of the candidate finds the corresponding road sign of the similar road sign of each candidate by inquiry table and believes
Breath；
(3) by shape similarity inspection, removal feature is similar and the road sign of shape dissmilarity, choose in candidate similar road sign and
Inquiry road sign feature is similar and the similar road sign of shape is as similar road sign.
10. the winding detection method as claimed in claim 8 based on the screening of road semanteme road sign, which is characterized in that the shape
Shape similarity inspection refers to:
For a certain inquiry road sign c and the similar road sign z of a candidate, it is considered phase if the two size meets following relationship
Seemingly:
Simultaneously
Wherein wc, hcRespectively indicate the width and height of road sign c, wz, wzRespectively indicate the width and height of road sign z.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811350086.5A CN109583329A (en) | 2018-11-13 | 2018-11-13 | Winding detection method based on the screening of road semanteme road sign |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811350086.5A CN109583329A (en) | 2018-11-13 | 2018-11-13 | Winding detection method based on the screening of road semanteme road sign |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109583329A true CN109583329A (en) | 2019-04-05 |
Family
ID=65922308
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811350086.5A Pending CN109583329A (en) | 2018-11-13 | 2018-11-13 | Winding detection method based on the screening of road semanteme road sign |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109583329A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110795994A (en) * | 2019-09-16 | 2020-02-14 | 腾讯科技（深圳）有限公司 | Intersection image selection method and device |
| CN110852353A (en) * | 2019-10-22 | 2020-02-28 | 上海眼控科技股份有限公司 | Intersection classification method and equipment |
| CN111780771A (en) * | 2020-05-12 | 2020-10-16 | 驭势科技（北京）有限公司 | Positioning method, positioning device, electronic equipment and computer readable storage medium |
| CN111986313A (en) * | 2020-08-21 | 2020-11-24 | 浙江商汤科技开发有限公司 | Loop detection method and device, electronic equipment and storage medium |
| CN118411542A (en) * | 2024-07-04 | 2024-07-30 | 南京智慧交通信息股份有限公司 | Road disease code generation and image matching verification method and system based on bus wheel inspection |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105841687A (en) * | 2015-01-14 | 2016-08-10 | 上海智乘网络科技有限公司 | Indoor location method and indoor location system |
| CN107833236A (en) * | 2017-10-31 | 2018-03-23 | 中国科学院电子学研究所 | Semantic vision positioning system and method are combined under a kind of dynamic environment |
| CN108596974A (en) * | 2018-04-04 | 2018-09-28 | 清华大学 | Dynamic scene robot localization builds drawing system and method |
-
2018
- 2018-11-13 CN CN201811350086.5A patent/CN109583329A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105841687A (en) * | 2015-01-14 | 2016-08-10 | 上海智乘网络科技有限公司 | Indoor location method and indoor location system |
| CN107833236A (en) * | 2017-10-31 | 2018-03-23 | 中国科学院电子学研究所 | Semantic vision positioning system and method are combined under a kind of dynamic environment |
| CN108596974A (en) * | 2018-04-04 | 2018-09-28 | 清华大学 | Dynamic scene robot localization builds drawing system and method |
Non-Patent Citations (4)
| Title |
|---|
| I_WOODEN: "计算两个矩形重叠面积的简单方法", 《HTTPS://BLOG.CSDN.NET/QIANCHENGLENGER/ARTICLE/DETAILS/50484053》 * |
| NIKO SÜNDERHAUF 等: "Place recognition with ConvNet landmarks: Viewpoint-robust condition-robust training-free", 《ROBOTICS: SCIENCE AND SYSTEMS CONFERENCE》 * |
| YI HOU 等: "Use of Roadway Scene Semantic Information and Geometry-Preserving Landmark Pairs to Improve Visual Place Recognition in Changing Environments", 《IEEE ACESS》 * |
| 杨恒 等: "面向高维图像特征匹配的多次随机子向量量化哈希算法", 《计算机辅助设计与图形学学报》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110795994A (en) * | 2019-09-16 | 2020-02-14 | 腾讯科技（深圳）有限公司 | Intersection image selection method and device |
| CN110852353A (en) * | 2019-10-22 | 2020-02-28 | 上海眼控科技股份有限公司 | Intersection classification method and equipment |
| CN111780771A (en) * | 2020-05-12 | 2020-10-16 | 驭势科技（北京）有限公司 | Positioning method, positioning device, electronic equipment and computer readable storage medium |
| CN111780771B (en) * | 2020-05-12 | 2022-09-23 | 驭势科技（北京）有限公司 | Positioning method, positioning device, electronic equipment and computer readable storage medium |
| CN111986313A (en) * | 2020-08-21 | 2020-11-24 | 浙江商汤科技开发有限公司 | Loop detection method and device, electronic equipment and storage medium |
| CN111986313B (en) * | 2020-08-21 | 2024-09-17 | 浙江商汤科技开发有限公司 | Loop detection method and device, electronic equipment and storage medium |
| CN118411542A (en) * | 2024-07-04 | 2024-07-30 | 南京智慧交通信息股份有限公司 | Road disease code generation and image matching verification method and system based on bus wheel inspection |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Lee et al. | Sliced wasserstein discrepancy for unsupervised domain adaptation | |
| Yuliang et al. | Detecting curve text in the wild: New dataset and new solution | |
| Jiang et al. | R2CNN: Rotational region CNN for orientation robust scene text detection | |
| CN109583329A (en) | Winding detection method based on the screening of road semanteme road sign | |
| Lei et al. | Region-enhanced convolutional neural network for object detection in remote sensing images | |
| CN110400332B (en) | Target detection tracking method and device and computer equipment | |
| JP6709283B2 (en) | Detection and analysis of moving vehicles using low resolution remote sensing images | |
| CN106682233B (en) | Hash image retrieval method based on deep learning and local feature fusion | |
| Wang et al. | Object detection using clustering algorithm adaptive searching regions in aerial images | |
| Lee et al. | Place recognition using straight lines for vision-based SLAM | |
| Li et al. | A complex junction recognition method based on GoogLeNet model | |
| Tran et al. | On-device scalable image-based localization via prioritized cascade search and fast one-many ransac | |
| Logoglu et al. | Cospair: colored histograms of spatial concentric surflet-pairs for 3d object recognition | |
| Li et al. | Place recognition based on deep feature and adaptive weighting of similarity matrix | |
| Cao et al. | FDTA: Fully convolutional scene text detection with text attention | |
| Fond et al. | Facade proposals for urban augmented reality | |
| Zhang et al. | Topological spatial verification for instance search | |
| Han et al. | Research on remote sensing image target recognition based on deep convolution neural network | |
| Mei et al. | Asymmetric global–local mutual integration network for RGBT tracking | |
| Chen et al. | A video-based method with strong-robustness for vehicle detection and classification based on static appearance features and motion features | |
| CN110246165B (en) | Method and system for improving registration speed of visible light image and SAR image | |
| Li et al. | Real-time tracking algorithm for aerial vehicles using improved convolutional neural network and transfer learning | |
| Ke et al. | Dense small face detection based on regional cascade multi‐scale method | |
| CN116778347A (en) | Data updating method, device, electronic equipment and storage medium | |
| CN116778346A (en) | Pipeline identification method and system based on improved self-attention mechanism |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190405 |
|
| RJ01 | Rejection of invention patent application after publication |