CN107220932A - Panorama Mosaic method based on bag of words - Google Patents

Panorama Mosaic method based on bag of words Download PDF

Info

Publication number
CN107220932A
CN107220932A CN201710251608.5A CN201710251608A CN107220932A CN 107220932 A CN107220932 A CN 107220932A CN 201710251608 A CN201710251608 A CN 201710251608A CN 107220932 A CN107220932 A CN 107220932A
Authority
CN
China
Prior art keywords
key frame
image
msub
mrow
images
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
Application number
CN201710251608.5A
Other languages
Chinese (zh)
Other versions
CN107220932B (en
Inventor
张国山
张玉龙
黄伟杰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tianjin University
Original Assignee
Tianjin University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Tianjin University filed Critical Tianjin University
Priority to CN201710251608.5A priority Critical patent/CN107220932B/en
Publication of CN107220932A publication Critical patent/CN107220932A/en
Application granted granted Critical
Publication of CN107220932B publication Critical patent/CN107220932B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
    • G06T3/4038—Image mosaicing, e.g. composing plane images from plane sub-images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/14—Transformations for image registration, e.g. adjusting or mapping for alignment of images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
    • G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
    • G06T7/337—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods involving reference images or patches

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Processing Or Creating Images (AREA)

Abstract

The present invention relates to a kind of Panorama Mosaic method based on bag of words, including：Prepare the ORB bag of words trees trained；For the image i in image data set to be spliced, ORB feature extractions are carried out, if k is upper one key frame for being inserted into splicing construction figure, signature search is carried out to two images by forward direction index, the correspondence set up between feature, so as to obtain the correspondence of feature between two images；Obtain the homography matrix between two images；Reprojection's error is minimized using random sampling maximum likelihood estimation algorithm and exterior point is excluded, and then obtains the interior point set of correspondence image pair；The overlapping percentages of image are calculated by boundary matrix；It is overlapping between calculating correspondence image pair, contacted when image i is added to as key frame in splicing construction figure, it is necessary to be set up with key frame before by homography matrix；Detect winding；Form winding；Optimize homography matrix；The fusion of image mosaic structure chart.

Description

Panorama Mosaic method based on bag of words
Technical field
The invention belongs to computer vision and digital image processing field.
Background technology
Image mosaic technology is an important branch of computer vision and digital image processing field, it be by two width with On that there is partly overlapping image to carry out is seamless spliced so as to obtaining the technology of high-resolution or wide viewing angle image.Image is spelled Two the most key links are image registration and image co-registration in connecing.For image fusion technology, at this stage method time-consuming and Difference less, has tended to be ripe in terms of syncretizing effect.But for image registration, because its registering time and effect are directly affected To the speed and success rate of image mosaic, so image registration is always the focus of present image splicing aspect research.Common Method for registering images is all based on SIFT either SURF feature description graphs as the advantage of these Feature Descriptors is to rotate Consistency, scale invariability and is difficult by illumination effect, and research at this stage begins to focus on BRIEF, ORB and BRISK etc. Binary features, because the less memory space of its needs and the rapidity of calculating.But image split-joint method at this stage Comparative approach of the frame to frame is all based on greatly, this method can receive good effect when picture number is less, still, with The increase of picture number, this method is not just applied to when requirement of real-time is higher.
The content of the invention
The purpose of the present invention is to propose to a kind of suitable for the higher panoramic picture spelling based on bag of words of requirement of real-time Method is connect, technical scheme is as follows.
A kind of Panorama Mosaic method based on bag of words, comprises the following steps：
1) the ORB bag of words trees trained are prepared；
2) for the image i in image data set to be spliced, ORB feature extractions is carried out, the feature extracted in image i is retouched State son and reach down to leaf node step by step since the root node of ORB bag of words trees according to Hamming distance, travel through all features Afterwards, the positive index of all features of the image is stored in bag of words tree, if k is upper one key for being inserted into splicing construction figure Two images are carried out signature search, the correspondence set up between feature, so that special between obtaining two images by frame by forward direction index The correspondence levied；
3) according to the correspondence of feature between obtained two images, the homography matrix between two images is obtainedkHi；
4) reprojection's error is minimized using random sampling maximum likelihood estimation algorithm and exterior point is excluded, and then obtained The interior point set of correspondence image pair；
5) boundary matrix of image is calculated according to obtained interior point set, the overlapping percentage of image is calculated by boundary matrix Than；
6) it is overlapping between calculating correspondence image pair according to the overlapping percentages of calculatingkOi=min (Ok,Oi), if interior The number of point is more than threshold tauinAnd it is overlappingkOiMore than threshold tauov, then image i is preserved as potential key frame, if treated Next two field picture is unsatisfactory for above-mentioned two threshold value with interior count out that key frame k is obtained with overlapping in splicing data set, then will figure As i as key frame is added to splicing construction figure；
7) when image i is added to as key frame in splicing construction figure, it is necessary to pass through homography with key frame before Matrix sets up contact, i.e. image i is the key frame of kth+1 of splicing construction figure, thenkHiIt is indicated askHk+1, then the key frame of kth+1 Homography matrixMHk+1It is expressed as：MHk+1=MHk kHk+1, wherein,MHkThe homography matrix of kth key frame is represented, M is for guarantor The general key frame demonstrate,proved the alignment between key frame and defined；
8) winding is detected, the key frame images i newly added is examined with key frame images all before splicing construction figure Rope is matched：The new ORB features for adding key frame images enter bag of words, according to Hamming distance since the root node of bag of words tree Leaf node is reached down to step by step, i.e. calculates the frequency that each word occurs in image i in each leaf node bag of words tree Tf, by the new all features for adding key frame images, retrieves in bag of words tree, obtains the value of each word, these values are constituted The description vectors of image；If it is respectively ν newly to add key frame images and the description vectors of matched previous keyframe image1 And ν2, the similarity score calculation formula of two width key frame images is expressed as：
Score is higher, and the similarity for representing this two width key frame images is higher, it is hereby achieved that newly adding key frame images With the similarity degree of key frame images before, so as to obtain a key frame similarity list from high to low, these are crucial Two field picture is to be possible to the key frame with newly adding key frame images formation winding.
9) according to key frame similarity tab sequential, these key frames and the new homography matrix for adding key frame are calculated, If by homography matrix obtain in count out the threshold value fixed more than one, then corresponding annexation, which just turns into, spells A part for binding composition, as forms winding；
10) optimize homography matrix, reduced using bundle adjustment as the error caused by homography matrix, error function ε is：
WhereinWithCharacter pair point in expression two images, R (MHi) it is homography matrixMHiRegular terms, be The influence of reduction exterior point, introducing Huber loss functions h (ε)=| ε |2if|ε|≤1；2 | ε | -1if | ε | ＞ 1 }, what is obtained is System nonlinear equation is solved by nonlinear least square method algorithm, so as to adjust optimization homography matrix；
11) fusion of image mosaic structure chart.
The main advantages of the present invention and characteristic be embodied in following aspects：
1st, image registration is all based on SIFT either SURF feature description graphs as benefiting in current merging algorithm for images In the yardstick of these features, rotational invariance and it is difficult by illumination effect, but the time mistake that the extraction of these features needs It is many, cause the real-time of algorithm not reach requirement, image registration in addition is the matching based on frame to frame, can also be increased with punctual Between.Index structure proposed by the present invention based on bag of words, using ORB Feature Descriptors.Experiment shows, based on bag of words Method for registering images while registration effect is similarly obtained, the algorithm time can be substantially reduced.
2nd, current merging algorithm for images is all based on having obvious order between the algorithm of single thread, algorithm various pieces greatly Property and coupling, algorithm of the invention can use multi-threaded architecture, be performed while realizing algorithm different piece, realize and protecting Demonstrate,prove on the basis of splicing effect, can effectively shorten the algorithm time.
Brief description of the drawings
Fig. 1 is the flow chart of the multithreading merging algorithm for images of the invention based on bag of words；
Fig. 2 is the image mosaic figure of Valldemossa data sets；
Fig. 3 is the topology diagram of Valldemossa data sets；
Fig. 4 is the image mosaic figure of Odemar data sets；
Fig. 5 is the topology diagram of Odemar data sets.
Embodiment
The present invention proposes the multithreading Panorama Mosaic technology based on bag of words, is described in detail with reference to example and accompanying drawing It is as follows：
The general frame of inventive algorithm is this as shown in figure 1, system is divided into four parts and this four part to run parallel Parallel design can reduce the coupling between various pieces, so as to reduce the time of algorithm operation.This four parts pass through One structure for being referred to as splicing construction figure is connected, and this structure is used for the topological structure for estimating to splice environment, while for assisting Adjust the relation between various pieces, it is ensured that real-time.
Splicing construction figure part is an important component of the inventive method, and topological diagram therein represents splicing ring The mechanism run between the topological structure in border, and unified different piece.The topological structure of environment represents to participate in image mosaic Image and its between contact.In the present invention, the mathematical modeling of topological structure is the form of non-directed graph, and wherein node on behalf Image selected in final splicing, connecting line represents the lap between them, in the present invention, selected image Referred to as key frame.In order to produce final spliced map, it is necessary to choose key frame, as stitching image frame.
The foundation of system other parts and splicing construction figure is synchronous progress, and key frame part describes input picture, entered Enter bag of words index structure processing image and determine that image is key frame, if be the part of final splicing construction image； Closed loop detection part can set up present frame in detection matching image to after and be contacted with matched key frame, form winding；It is excellent Change part and homography matrix is adjusted by bundle adjustment, to reduce the error that error hiding is caused；Splicing construction after optimization Figure enters fusion part and produces last spliced map.Specific embodiment is as follows：
7) bag of words tree is built, DBoW2 Cooleys have got well ORB storehouses and SIFT storehouses with a big image data base, off-line training, Used for everybody.In the present invention, the ORB bag of words trees trained in DBoW2 storehouses are used.
8) for the image i in image data set to be spliced, ORB feature extractions is carried out, the feature extracted in image i is retouched State son and reach down to leaf node step by step since the root node of ORB bag of words trees according to Hamming distance, travel through all features Afterwards, the positive index of all features of the image is stored in bag of words tree.K is upper one key frame for being inserted into splicing construction figure, Signature search, the correspondence set up between feature, so as to obtain feature between two images are carried out to two images by forward direction index Correspondence；
9) according to the correspondence of feature between obtained two images, the homography matrix between two images is obtainedkHi；
10) reprojection's error is minimized using random sampling maximum likelihood estimation algorithm and exterior point is excluded, and then To the interior point set of correspondence image pair；
11) boundary matrix of image is calculated according to obtained interior point set, the overlapping percentage of image is calculated by boundary matrix Than；
12) it is overlapping between calculating correspondence image pair according to the overlapping percentages of calculatingkOi=min (Ok,Oi), if interior The number of point is more than threshold tauinAnd it is overlappingkOiMore than threshold tauov, then image i is preserved as potential key frame, if treated Next two field picture is unsatisfactory for above-mentioned two threshold value with interior count out that key frame k is obtained with overlapping in splicing data set, then will figure As i as key frame is added to splicing construction figure；
7) when image i is added to as key frame in splicing construction figure, it is necessary to pass through homography with key frame before Matrix sets up contact, then the homography matrix of the key frame of kth+1MHk+1It is expressed as：MHk+1=MHk kHk+1, wherein M is to ensure Alignment between key frame and in the general key frame defined, splicing construction figure each key frame homography matrixMHk+1,MHkAll it is to be set up with key frame M；
8) winding is detected, the key frame images i newly added is examined with key frame images all before splicing construction figure Rope is matched：The new ORB features for adding key frame images enter bag of words, according to Hamming distance since the root node of bag of words tree Leaf node is reached down to step by step, i.e. calculates the frequency that each word occurs in image i in each leaf node bag of words tree Tf, each leaf node stores reverse indexing in bag of words tree, as stores the image ID and word for reaching leaf node Value in iamge description vector, by the new all features for adding key frame images, retrieves in bag of words tree, obtains each word Value, by the description vectors of these value pie graph pictures；If newly adding key frame images and matched previous keyframe image Description vectors be respectively ν1And ν2, the similarity score calculation formula of two width key frame images is expressed as：
Score is higher, and the similarity for representing this two width key frame images is higher, it is hereby achieved that newly adding key frame images With key frame figure before
The similarity degree of picture, so as to obtain a key frame similarity list from high to low, these key frame images As it is possible to close with newly adding
The key frame of key two field picture formation winding.
9) according to key frame similarity tab sequential, these key frames and the new homography matrix for adding key frame are calculated, If by homography matrix obtain in count out the threshold value fixed more than one, then corresponding annexation, which just turns into, spells A part for binding composition, as forms winding；
10) optimize and all contacted between homography matrix, key frame images by homography matrix, but list should Property matrix there is error, it is necessary to optimize, using bundle adjustment reduce as the error caused by homography matrix, error Function of ε is：
WhereinWithCharacter pair point in expression two images, R (MHi) it is homography matrixMHiRegular terms, be Reduce the influence of exterior point, introduce Huber loss functions h (ε)=| ε |2if|ε|≤1；2 | ε | -1if | ε | ＞ 1 }；Obtain Mission nonlinear equation is solved by nonlinear least square method algorithm, and homography matrix is as initial value, by iteration just It can be restrained, so as to adjust homography matrix.
11) fusion of image mosaic structure chart.
Fusion is the final step of stitching algorithm, and the seamless spliced figure last for producing, this part is OPENCV storehouses Middle stitching application, including seam and line technology and exposure compensating, in the present invention, this part are produced before Splicing construction figure can form final seamless spliced figure, i.e., final experiment effect figure immediately.
In order to verify the validity and real-time of the inventive method, the present invention selects two group data sets.Valldemossa numbers According to collecting, this dataset acquisition includes 201 320 × 180 in Hispanic port city Valldemossa underwater environment Picture, is to be overlooked to shoot by camera, this data set includes a big closed loop；Odemar data sets, this data set is served as reasons The underwater environment of Miquel Massot-Campos collections, includes the picture of 64 480 × 270, is to look up shooting by camera, This data set does not include big closed loop.Experimental result is as follows：
1. the spliced map of obtained Valldemossa data sets is as shown in Fig. 2 the topology estimation of environment is as shown in figure 3, open up Flutter figure and contain 76 frame key frames.The spliced map of Odemar data sets is obtained as shown in figure 4, topology diagram such as Fig. 5 institutes of environment Show, topological diagram contains 22 frame key frames.
2. this group experiment is contrast test, method and the ratio of the single thread joining method using ORB features by the present invention Compared with still from this two group data set, obtained experimental data is as shown in table 1, it can be seen that in reprojection's error difference not In the case of big, the time of algorithm is greatly improved.
Table 1 is data comparison of the experimental data with the single thread merging algorithm for images based on ORB features of the present invention；
Table 1.

Claims (1)

1. a kind of Panorama Mosaic method based on bag of words, comprises the following steps：
1) the ORB bag of words trees trained are prepared；
2) for the image i in image data set to be spliced, ORB feature extractions, the Feature Descriptor that will be extracted in image i are carried out Leaf node is reached down to step by step since the root node of ORB bag of words trees according to Hamming distance, after having traveled through all features, The positive index of all features of the image is stored in bag of words tree, if k is upper one key frame for being inserted into splicing construction figure, is passed through Forward direction index carries out signature search, the correspondence set up between feature, so as to obtain pair of feature between two images to two images Should；
3) according to the correspondence of feature between obtained two images, the homography matrix between two images is obtainedkHi；
4) reprojection's error is minimized using random sampling maximum likelihood estimation algorithm and exterior point is excluded, and then obtains correspondence The interior point set of image pair；
5) boundary matrix of image is calculated according to obtained interior point set, the overlapping percentages of image are calculated by boundary matrix；
6) according to the overlapping percentages of calculating, the overlapping kO between correspondence image pair is calculatedi=min (Ok,Oi), if interior point Number is more than threshold tauinAnd it is overlappingkOiMore than threshold tauov, then image i is preserved as potential key frame, if to be spliced Next two field picture is unsatisfactory for above-mentioned two threshold value with interior count out that key frame k is obtained with overlapping in data set, then makees image i Splicing construction figure is added to for key frame；
7) when image i is added to as key frame in splicing construction figure, it is necessary to pass through homography matrix with key frame before Contact is set up, i.e. image i is the key frame of kth+1 of splicing construction figure, thenkHiIt is indicated askHk+1, then the list of the key frame of kth+1 Answering property matrixMHk+1It is expressed as：MHk+1=MHk kHk+1, wherein,MHkThe homography matrix of kth key frame is represented, M is closed for guarantee Alignment between key frame and the general key frame defined；
8) winding is detected, the key frame images i newly added carries out retrieval with key frame images all before splicing construction figure Match somebody with somebody：The new ORB features for adding key frame images enter bag of words, according to Hamming distance since the root node of bag of words tree step by step Leaf node is reached down to, i.e. calculates the frequency tf that each word occurs in image i in each leaf node bag of words tree, will The new all features for adding key frame images, retrieve in bag of words tree, obtain the value of each word, by these value pie graphs as Description vectors；If it is respectively ν newly to add key frame images and the description vectors of matched previous keyframe image1And ν2, two The similarity score calculation formula of width key frame images is expressed as：
<mrow> <mi>s</mi> <mrow> <mo>(</mo> <msub> <mi>&amp;nu;</mi> <mn>1</mn> </msub> <mo>,</mo> <msub> <mi>&amp;nu;</mi> <mn>2</mn> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mn>1</mn> <mo>-</mo> <mfrac> <mn>1</mn> <mn>2</mn> </mfrac> <mo>|</mo> <mfrac> <msub> <mi>&amp;nu;</mi> <mn>1</mn> </msub> <mrow> <mo>|</mo> <msub> <mi>&amp;nu;</mi> <mn>1</mn> </msub> <mo>|</mo> </mrow> </mfrac> <mo>-</mo> <mfrac> <msub> <mi>&amp;nu;</mi> <mn>2</mn> </msub> <mrow> <mo>|</mo> <msub> <mi>&amp;nu;</mi> <mn>2</mn> </msub> <mo>|</mo> </mrow> </mfrac> <mo>|</mo> </mrow>
Score is higher, and the similarity for representing this two width key frame images is higher, it is hereby achieved that newly adding key frame images therewith The similarity degree of preceding key frame images, so as to obtain a key frame similarity list from high to low, these key frame figures As being to be possible to the key frame with newly adding key frame images formation winding.
9) according to key frame similarity tab sequential, these key frames and the new homography matrix for adding key frame are calculated, if By homography matrix obtain in count out the threshold value fixed more than one, then corresponding annexation just turns into splicing knot A part for composition, as forms winding；
10) optimize homography matrix, reduced using bundle adjustment as the error caused by homography matrix, error function ε is：
<mrow> <mi>&amp;epsiv;</mi> <mo>=</mo> <munder> <mo>&amp;Sigma;</mo> <mi>i</mi> </munder> <munder> <mo>&amp;Sigma;</mo> <mi>j</mi> </munder> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>k</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>n</mi> </munderover> <mo>|</mo> <mo>|</mo> <msubsup> <mi>p</mi> <mi>i</mi> <mi>k</mi> </msubsup> <mo>-</mo> <msup> <mrow> <mo>(</mo> <msub> <mmultiscripts> <mi>H</mi> <mi>M</mi> </mmultiscripts> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mrow> <mo>-</mo> <mn>1</mn> </mrow> </msup> <msub> <mmultiscripts> <mi>H</mi> <mi>M</mi> </mmultiscripts> <mi>j</mi> </msub> <msubsup> <mi>p</mi> <mi>j</mi> <mi>k</mi> </msubsup> <mo>|</mo> <mo>|</mo> <mo>+</mo> <mi>R</mi> <mrow> <mo>(</mo> <msub> <mmultiscripts> <mi>H</mi> <mi>M</mi> </mmultiscripts> <mi>j</mi> </msub> <mo>)</mo> </mrow> <mo>|</mo> <mo>|</mo> <msubsup> <mi>p</mi> <mi>j</mi> <mi>k</mi> </msubsup> <mo>-</mo> <msup> <mrow> <mo>(</mo> <msub> <mmultiscripts> <mi>H</mi> <mi>M</mi> </mmultiscripts> <mi>j</mi> </msub> <mo>)</mo> </mrow> <mrow> <mo>-</mo> <mn>1</mn> </mrow> </msup> <msub> <mmultiscripts> <mi>H</mi> <mi>M</mi> </mmultiscripts> <mi>i</mi> </msub> <msubsup> <mi>p</mi> <mi>i</mi> <mi>k</mi> </msubsup> <mo>|</mo> <mo>|</mo> <mo>+</mo> <mi>R</mi> <mrow> <mo>(</mo> <msub> <mmultiscripts> <mi>H</mi> <mi>M</mi> </mmultiscripts> <mi>i</mi> </msub> <mo>)</mo> </mrow> </mrow>
WhereinWithCharacter pair point in expression two images, R (MHi) it is homography matrixMHiRegular terms, for reduce The influence of exterior point, introducing Huber loss functions h (ε)=| ε |2if|ε|≤1；2 | ε | -1if | ε | ＞ 1 }, obtained system is non- Linear equation is solved by nonlinear least square method algorithm, so as to adjust optimization homography matrix；
11) fusion of image mosaic structure chart.
CN201710251608.5A 2017-04-18 2017-04-18 Panoramic image splicing method based on bag-of-words model Active CN107220932B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710251608.5A CN107220932B (en) 2017-04-18 2017-04-18 Panoramic image splicing method based on bag-of-words model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710251608.5A CN107220932B (en) 2017-04-18 2017-04-18 Panoramic image splicing method based on bag-of-words model

Publications (2)

Publication Number Publication Date
CN107220932A true CN107220932A (en) 2017-09-29
CN107220932B CN107220932B (en) 2020-03-20

Family

ID=59927589

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710251608.5A Active CN107220932B (en) 2017-04-18 2017-04-18 Panoramic image splicing method based on bag-of-words model

Country Status (1)

Country Link
CN (1) CN107220932B (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109101981A (en) * 2018-07-19 2018-12-28 东南大学 Winding detection method based on global image bar code under a kind of streetscape scene
CN109376631A (en) * 2018-10-12 2019-02-22 中国人民公安大学 A kind of winding detection method and device neural network based
CN109579844A (en) * 2018-12-04 2019-04-05 电子科技大学 Localization method and system
CN116934591A (en) * 2023-06-28 2023-10-24 深圳市碧云祥电子有限公司 Image stitching method, device and equipment for multi-scale feature extraction and storage medium

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105856230A (en) * 2016-05-06 2016-08-17 简燕梅 ORB key frame closed-loop detection SLAM method capable of improving consistency of position and pose of robot
CN106529583A (en) * 2016-11-01 2017-03-22 哈尔滨工程大学 Bag-of-visual-word-model-based indoor scene cognitive method

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105856230A (en) * 2016-05-06 2016-08-17 简燕梅 ORB key frame closed-loop detection SLAM method capable of improving consistency of position and pose of robot
CN106529583A (en) * 2016-11-01 2017-03-22 哈尔滨工程大学 Bag-of-visual-word-model-based indoor scene cognitive method

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
FANG TIAN等: "Image Mosaic using ORB descriptor and improved blending algorithm", 《2014 7TH INTERNATIONAL CONGRESS ON IMAGE AND SIGNAL PROCESSING》 *
余杰: "基于ORB关键帧闭环检测算法的SLAM方法研究", 《中国学位论文全文数据库》 *
姜鹏飞: "基于ORB特征的视频图像拼接技术研究", 《《中国优秀硕士学位论文全文数据库信息科技辑》》 *

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109101981A (en) * 2018-07-19 2018-12-28 东南大学 Winding detection method based on global image bar code under a kind of streetscape scene
CN109101981B (en) * 2018-07-19 2021-08-24 东南大学 Loop detection method based on global image stripe code in streetscape scene
CN109376631A (en) * 2018-10-12 2019-02-22 中国人民公安大学 A kind of winding detection method and device neural network based
CN109579844A (en) * 2018-12-04 2019-04-05 电子科技大学 Localization method and system
CN109579844B (en) * 2018-12-04 2023-11-21 电子科技大学 Positioning method and system
CN116934591A (en) * 2023-06-28 2023-10-24 深圳市碧云祥电子有限公司 Image stitching method, device and equipment for multi-scale feature extraction and storage medium

Also Published As

Publication number Publication date
CN107220932B (en) 2020-03-20

Similar Documents

Publication Publication Date Title
Anwar et al. Image colorization: A survey and dataset
Fan et al. Dual refinement underwater object detection network
CN107291945B (en) High-precision clothing image retrieval method and system based on visual attention model
CN111311666B (en) Monocular vision odometer method integrating edge features and deep learning
CN107220932A (en) Panorama Mosaic method based on bag of words
CN106534616B (en) A kind of video image stabilization method and system based on characteristic matching and motion compensation
CN107833179A (en) The quick joining method and system of a kind of infrared image
CN109829853B (en) Unmanned aerial vehicle aerial image splicing method
CN107424181A (en) A kind of improved image mosaic key frame rapid extracting method
US7936915B2 (en) Focal length estimation for panoramic stitching
CN107767419A (en) A kind of skeleton critical point detection method and device
CN106651767A (en) Panoramic image obtaining method and apparatus
CN107316275A (en) A kind of large scale Microscopic Image Mosaicing algorithm of light stream auxiliary
CN105761233A (en) FPGA-based real-time panoramic image mosaic method
CN104616247B (en) A kind of method for map splicing of being taken photo by plane based on super-pixel SIFT
CN110097498B (en) Multi-flight-zone image splicing and positioning method based on unmanned aerial vehicle flight path constraint
CN114418853B (en) Image super-resolution optimization method, medium and equipment based on similar image retrieval
WO2023066173A1 (en) Image processing method and apparatus, and storage medium and electronic device
CN113724135A (en) Image splicing method, device, equipment and storage medium
CN106530266A (en) Infrared and visible light image fusion method based on area sparse representation
CN111815690B (en) Method, system and computer equipment for real-time splicing of microscopic images
Hughes et al. A semi-supervised approach to SAR-optical image matching
CN109034237A (en) Winding detection method based on convolutional Neural metanetwork road sign and sequence search
CN109598675B (en) Splicing method of multiple repeated texture images
CN116523980A (en) Cross-modal image registration method based on test phase optimization

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