CN112435296A - Image matching method for VSLAM indoor high-precision positioning - Google Patents
Image matching method for VSLAM indoor high-precision positioning Download PDFInfo
- Publication number
- CN112435296A CN112435296A CN202011380479.8A CN202011380479A CN112435296A CN 112435296 A CN112435296 A CN 112435296A CN 202011380479 A CN202011380479 A CN 202011380479A CN 112435296 A CN112435296 A CN 112435296A
- Authority
- CN
- China
- Prior art keywords
- value
- matching
- image
- illumination
- vslam
- 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
- 238000000034 method Methods 0.000 title claims abstract description 21
- 238000005286 illumination Methods 0.000 claims abstract description 19
- 238000005259 measurement Methods 0.000 claims abstract description 8
- 230000001629 suppression Effects 0.000 claims abstract description 4
- 230000005764 inhibitory process Effects 0.000 claims description 10
- 230000019935 photoinhibition Effects 0.000 claims description 3
- 230000004807 localization Effects 0.000 claims 1
- 238000001514 detection method Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000007670 refining Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
- G06T7/74—Determining position or orientation of objects or cameras using feature-based methods involving reference images or patches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
- G06T7/248—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments involving reference images or patches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30241—Trajectory
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30244—Camera pose
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to an image matching method for VSLAM indoor high-precision positioning, which can better detect repeatable key points under severe illumination change conditions by adding an illumination suppression factor in a matching similarity criterion, has illumination robustness and can improve the accuracy of image matching. In addition, the invention also comprehensively utilizes cosine similarity and Euclidean measurement, can measure the difference between the value direction of the key point and the absolute value in the image matching process, and further improves the accuracy of image matching.
Description
Technical Field
The invention relates to an image matching method for VSLAM indoor high-precision positioning.
Background
The vision SLAM system is mainly applied to robot vision positioning, and the working process includes front end image acquisition, rear end matching optimization and closed loop detection etc. and the image matching includes: extracting feature points of each frame of image; and carrying out image matching by using the characteristic point coordinates and the descriptors to obtain the motion trail of the camera. The method comprises the steps of calculating the matching relation between a current frame image and an adjacent frame image, solving the relative spatial relation between two scenes, and accordingly obtaining the motion trail of a camera is a key link.
On the other hand, in the image matching process, the euclidean distance or an improved algorithm thereof is a common similarity discrimination criterion, the euclidean distance mainly discriminates the absolute difference of the real distance between two points in the m-dimensional space, and the cosine similarity more discriminates the difference from the m-dimensional direction and is insensitive to an absolute numerical value.
Disclosure of Invention
The inventor finds that in the process of photographing or video recording by a depth camera, due to unpredictability of day and night periods, weather changes and lighting environments, dynamic changes can occur in lighting conditions, the accuracy of extracting feature points of an image can be influenced under the condition of severe lighting changes, and many problems exist in the loop detection and back-end optimization links for identifying and matching the feature points of the image under two lighting conditions with great differences, so that failure in positioning or reduction in positioning accuracy can be caused easily.
In addition, in the image matching process, only the numerical difference of the feature points or the direction vector difference of the feature points are considered, which affects the accuracy of the image matching.
Aiming at the problems, the invention provides the image matching method for VSLAM indoor high-precision positioning, which has good illumination robustness and high image matching accuracy.
The technical scheme adopted by the invention is as follows:
an image matching method for VSLAM indoor high-precision positioning comprises the following steps:
step one, extracting feature points of each frame of image;
step two, establishing a matching similarity criterion, combining Euclidean measurement and cosine similarity measurement in the matching similarity criterion, and adding an illumination inhibition factor; the matching similarity criterion function E is:
in the formula (1), both alpha and beta are numeric coefficients, and the numeric range is (0, 1); sqrt represents taking the positive square root; gamma (t) is an illumination inhibition factor, and the value range is (0, 1); w is aiThe value range is (0,1) for the weighting coefficient; x is the number ofi1Representing a first feature pointOf the ith dimension, xi2An ith coordinate representing the second feature point, i ═ 1,2, … n;
the photoinhibition factor γ (t) is expressed as:
in the formula (2), eta is an inhibition coefficient, and the numeric area is (0, 1); Δ i (t) is a change value of the luminous intensity of the light source with time, and r is a distance from the camera to the light source;
step three, calculating a matching similarity metric value E of the images of the adjacent frames through a matching similarity criterion function, sorting the images according to the calculated similarity metric values from large to small, and selecting the image with the smallest value E as a matching frame;
and step four, quantitatively estimating the motion of the camera between frames according to the motion vector change between the adjacent matched frames to obtain the position and the posture of the camera.
Further, in the third step, when the image positioning accuracy exceeds 1m, it is determined that the illumination condition is changed drastically, and the value step length of the illumination suppression factor γ (t) is dynamically adjusted within the range of (0, 1); otherwise, the possible value of each alpha and beta parameter is calculated by adjusting the alpha and beta value step length in the range of (0,1), and then all the combination conditions are traversed and the optimal parameter value is returned.
Further, β is set to 0.9, α is set to 0.1; alternatively, the initial value of β is set to 0.1 and the initial value of α is set to 0.9.
The invention has the beneficial effects that:
according to the method, the illumination suppression factor is added in the matching similarity criterion, so that repeatable key points under the condition of severe illumination change can be well detected, the illumination robustness is achieved, and the accuracy of image matching can be improved. In addition, the invention also comprehensively utilizes cosine similarity and Euclidean measurement, can measure the difference (the difference in space and distance) between the value direction and the absolute value of the key point in the image matching process, and further improves the accuracy of image matching.
Drawings
Fig. 1 is a flow chart of the image matching method for VSLAM indoor high-precision positioning according to the present invention.
Detailed Description
The image matching method for high-precision positioning of the VSLAM in the room according to the present invention will be further described with reference to the accompanying drawings and specific embodiments.
As shown in fig. 1, an image matching method for high-precision positioning of VSLAM in a room includes the following steps:
step one, extracting feature points (including coordinate information and descriptors and direction information on 8 dimensions) of each frame of image.
And step two, establishing a matching similarity criterion, combining Euclidean measurement and cosine similarity measurement in the matching similarity criterion, and adding an illumination inhibition factor. The matching similarity criterion function E is:
in the formula (1), both α and β are numeric coefficients, and the numeric range is (0, 1). sqrt denotes taking the positive square root. Gamma (t) is an illumination inhibition factor, and the value range is (0, 1). w is aiThe value range is (0,1) for the weighting coefficient. x is the number ofi1I-dimensional coordinate, x, representing the first feature pointi2And i-th-dimension coordinates representing the second feature point, i being 1,2, … n.
The photoinhibition factor γ (t) is expressed as:
in the formula (2), η is an inhibition coefficient, and the numeric area is (0, 1). Δ I (t) is the variation value of the luminous intensity of the light source with time, and r is the distance from the camera to the light source.
In the matching similarity criterion of the invention, the difference judgment of the absolute distances of the feature points is considered, and the Euclidean measure d is utilized. And (4) considering the influence of the feature point direction dimension information, and measuring cos by using cosine similarity.
d＝sqrt(∑(xi1-xi2)2) (3)
The euclidean metric d measures the absolute distance between feature points.
The closer the cosine value is to 1, the closer the included angle is to 0 degrees, i.e., the more similar the two feature points are.
And step three, calculating a matching similarity metric value E of the images of the adjacent frames through a matching similarity criterion function, sorting the images according to the calculated similarity metric values from large to small, and selecting the image with the smallest value E as a matching frame.
And when the image quality does not meet the requirement and the positioning accuracy is poor (exceeds 1m), judging that the illumination condition is changed violently, and dynamically adjusting the value step length of the illumination inhibition factor gamma (t) in the range of (0, 1). Otherwise, the possible value of each alpha and beta parameter is calculated by adjusting the alpha and beta value step length in the range of (0,1), and then all the combination conditions are traversed and the optimal parameter value is returned.
Beta reflects the distance penalty coefficient of the model to the matching criterion, and alpha reflects the distribution of the data after being mapped to the high-dimensional feature space. The larger β, the easier the model is to overfit. The smaller beta, the easier the model is to under-fit. The larger alpha is, the more supporting feature point descriptor directions are, the smaller alpha value is, the fewer supporting feature point descriptor directions are. The smaller α is, the better the generalization of the model becomes, but too small, the model is degraded into a linear model in practice. The larger the α, the theoretically possible to fit any non-linear data.
Before the simulation starts, the value of alpha is set, the value of beta is set to be 0.1-1, the optimal parameter range is that the initial value of beta is set to be 0.9, and the initial value of alpha is set to be 0.1. Alternatively, the initial value of β is set to 0.1 and the initial value of α is set to 0.9. Then, according to the representation of the model precision, increasing beta or increasing alpha (not increasing simultaneously), or needing to decrease beta or decreasing alpha, namely multiplying 0.1 or 10 each time as a step length, and after determining the approximate range, refining the search matching interval.
And step four, quantitatively estimating the motion of the camera between frames according to the motion vector change between the adjacent matched frames to obtain the position and the posture of the camera.
The above description is only an embodiment of the present invention, but the scope of the present invention is not limited thereto, and any alternative or alternative method that can be easily conceived by those skilled in the art within the technical scope of the present invention should be covered by the scope of the present invention.
Claims (3)
1. An image matching method for VSLAM indoor high-precision positioning is characterized by comprising the following steps:
step one, extracting feature points of each frame of image;
step two, establishing a matching similarity criterion, combining Euclidean measurement and cosine similarity measurement in the matching similarity criterion, and adding an illumination inhibition factor; the matching similarity criterion function E is:
in the formula (1), both alpha and beta are numeric coefficients, and the numeric range is (0, 1); sqrt represents taking the positive square root; gamma (t) is an illumination inhibition factor, and the value range is (0, 1); w is aiThe value range is (0,1) for the weighting coefficient; x is the number ofi1I-dimensional coordinate, x, representing the first feature pointi2An ith coordinate representing the second feature point, i ═ 1,2, … n;
the photoinhibition factor γ (t) is expressed as:
in the formula (2), eta is an inhibition coefficient, and the numeric area is (0, 1); Δ i (t) is a change value of the luminous intensity of the light source with time, and r is a distance from the camera to the light source;
step three, calculating a matching similarity metric value E of the images of the adjacent frames through a matching similarity criterion function, sorting the images according to the calculated similarity metric values from large to small, and selecting the image with the smallest value E as a matching frame;
and step four, quantitatively estimating the motion of the camera between frames according to the motion vector change between the adjacent matched frames to obtain the position and the posture of the camera.
2. The image matching method for VSLAM indoor high-precision positioning according to claim 1, wherein in step three, when the image positioning precision exceeds 1m, it is determined that the illumination condition has changed drastically, and the value step length of the illumination suppression factor γ (t) is dynamically adjusted within the range of (0, 1); otherwise, the possible value of each alpha and beta parameter is calculated by adjusting the alpha and beta value step length in the range of (0,1), and then all the combination conditions are traversed and the optimal parameter value is returned.
3. The image matching method for VSLAM high accuracy localization in a room of claim 2, wherein β initial value is set to 0.9, α initial value is set to 0.1; alternatively, the initial value of β is set to 0.1 and the initial value of α is set to 0.9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011380479.8A CN112435296B (en) | 2020-12-01 | 2020-12-01 | Image matching method for VSLAM indoor high-precision positioning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011380479.8A CN112435296B (en) | 2020-12-01 | 2020-12-01 | Image matching method for VSLAM indoor high-precision positioning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112435296A true CN112435296A (en) | 2021-03-02 |
| CN112435296B CN112435296B (en) | 2024-04-19 |
Family
ID=74699003
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011380479.8A Active CN112435296B (en) | 2020-12-01 | 2020-12-01 | Image matching method for VSLAM indoor high-precision positioning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112435296B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2023044661A1 (en) * | 2021-09-23 | 2023-03-30 | Intel Corporation | Learning reliable keypoints in situ with introspective self-supervision |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107330357A (en) * | 2017-05-18 | 2017-11-07 | 东北大学 | Vision SLAM closed loop detection methods based on deep neural network |
| US20180297207A1 (en) * | 2017-04-14 | 2018-10-18 | TwoAntz, Inc. | Visual positioning and navigation device and method thereof |
| CN110097093A (en) * | 2019-04-15 | 2019-08-06 | 河海大学 | A kind of heterologous accurate matching of image method |
| CN111797938A (en) * | 2020-07-15 | 2020-10-20 | 燕山大学 | Semantic information and VSLAM fusion method for sweeping robot |
-
2020
- 2020-12-01 CN CN202011380479.8A patent/CN112435296B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180297207A1 (en) * | 2017-04-14 | 2018-10-18 | TwoAntz, Inc. | Visual positioning and navigation device and method thereof |
| CN107330357A (en) * | 2017-05-18 | 2017-11-07 | 东北大学 | Vision SLAM closed loop detection methods based on deep neural network |
| CN110097093A (en) * | 2019-04-15 | 2019-08-06 | 河海大学 | A kind of heterologous accurate matching of image method |
| CN111797938A (en) * | 2020-07-15 | 2020-10-20 | 燕山大学 | Semantic information and VSLAM fusion method for sweeping robot |
Non-Patent Citations (2)
| Title |
|---|
| MING FAN, SEUNG-WON JUNG, SUNG-JEA KO: "Highly Accurate Scale Estimation from Multiple Keyframes Using RANSAC Plane Fitting With a Novel Scoring Method", IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY ( VOLUME: 69, ISSUE: 12, DECEMBER 2020), 24 November 2020 (2020-11-24) * |
| 李博: "基于场景外观建模的移动机器人视觉闭环检测研究", 中国博士论文全文数据库, 15 December 2011 (2011-12-15) * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2023044661A1 (en) * | 2021-09-23 | 2023-03-30 | Intel Corporation | Learning reliable keypoints in situ with introspective self-supervision |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112435296B (en) | 2024-04-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN1871622B (en) | Image collation system and image collation method | |
| CN110807809B (en) | Light-weight monocular vision positioning method based on point-line characteristics and depth filter | |
| Ellis | Performance metrics and methods for tracking in surveillance | |
| US7706601B2 (en) | Object posture estimation/correlation system using weight information | |
| CN106682603B (en) | Real-time driver fatigue early warning system based on multi-source information fusion | |
| CN110889324A (en) | Thermal infrared image target identification method based on YOLO V3 terminal-oriented guidance | |
| CN111476827B (en) | Target tracking method, system, electronic device and storage medium | |
| CN113554125B (en) | Object detection apparatus, method and storage medium combining global and local features | |
| CN108197604A (en) | Fast face positioning and tracing method based on embedded device | |
| CN107403451B (en) | Self-adaptive binary characteristic monocular vision odometer method, computer and robot | |
| CN107622502A (en) | The path extraction of robot vision leading system and recognition methods under the conditions of complex illumination | |
| CN101777129A (en) | Image matching method based on feature detection | |
| CN108871290B (en) | visible light dynamic positioning method based on optical flow detection and Bayesian prediction | |
| CN103093458A (en) | Detecting method and detecting device for key frame | |
| CN116363694A (en) | Multi-target tracking method of unmanned system crossing cameras matched with multiple pieces of information | |
| CN112435296A (en) | Image matching method for VSLAM indoor high-precision positioning | |
| CN113362377B (en) | VO weighted optimization method based on monocular camera | |
| Savinykh et al. | Darkslam: Gan-assisted visual slam for reliable operation in low-light conditions | |
| CN111476314B (en) | Fuzzy video detection method integrating optical flow algorithm and deep learning | |
| CN109064497B (en) | Video tracking method based on color clustering supplementary learning | |
| CN111161323A (en) | Complex scene target tracking method and system based on correlation filtering | |
| JP2004054947A (en) | Object collation system, object collation method and object collation program | |
| Peltomäki et al. | Evaluation of long-term LiDAR place recognition | |
| CN116403085A (en) | Multi-pointer instrument identification method and device for transformer substation based on multitasking | |
| CN115359124A (en) | High-precision binocular vision measuring method and device used under complex environment condition |
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 |