CN108022263A - A kind of SIFT feature inspection optimization method based on the notable parameter index of regional area - Google Patents
A kind of SIFT feature inspection optimization method based on the notable parameter index of regional area Download PDFInfo
- Publication number
- CN108022263A CN108022263A CN201711264919.1A CN201711264919A CN108022263A CN 108022263 A CN108022263 A CN 108022263A CN 201711264919 A CN201711264919 A CN 201711264919A CN 108022263 A CN108022263 A CN 108022263A
- Authority
- CN
- China
- Prior art keywords
- characteristic point
- sift feature
- regional area
- feature
- point
- 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
- 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
-
- 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
- G06T2207/10004—Still image; Photographic image
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of SIFT feature inspection optimization method based on the notable parameter index of regional area, including:Detect corresponding SIFT feature;Image is divided into regional area, chooses local marking area;For each local marking area, therefrom selected characteristic point;Selected SIFT feature is matched with current signature map.The advantage of the invention is that:It is more advantageous to improving the convergence rate of monocular SLAM systems, while also contributes to more accurately describe scene environment.
Description
Technical field
It is more particularly to a kind of to be based on the notable parameter index of regional area the present invention relates to positioning and map structuring technical field
SIFT feature inspection optimization method.
Background technology
The synchronous positioning of mobile robot and map building (Simultaneous Localization under uncertain environment
And Mapping, SLAM) it is one of key issue extremely challenging in current robot positioning and navigation research field.
The solution of SLAM problems is that robot realizes height autonomy-oriented and intelligentized premise.In recent years, with computer vision technique
Development and visual sensor extensive use, the research of monocular vision SLAM technologies has been increasingly becoming in SLAM research fields
Important research direction.What the SLAM methods of view-based access control model had to face first is exactly the extraction problem of visual signature.Vision
SLAM methods must obtain in image stablize and write domestic animal characteristic information, how from numerous visual information required feature
Information fully effectively propose be vision SLAM research hot spot and difficult point.
Restricted by vision technique and computer performance, the current progress in relation to machine vision is not also too fast.Pin
To some complicated unknown scenes, robot can not as people moment perceive and the situation of surrounding environment residing for capturing, institute
It is a difficulties in monocular SLAM with the observation road sign which kind of scene characteristic extracted as robot.Due to regarding
Feel that the primary data information that sensor obtains is image, therefore figure must be just used to construct the map description of residing scene
As preconditioning technique and feature extracting method therefrom extract required information amount.At present, the feature that can be extracted from visual pattern
Species is more, common to have point, line, surface, profile etc..Using point feature detective operators come to obtain scene characteristic point be still current list
The usual way that map feature is extracted in mesh SLAM Study on Problems.
Natural scene map road sign is generally obtained using SIFT feature extraction algorithm in vision SLAM.Scale invariant feature
Conversion (Scale Invariant Feature Transform, SIFT) is to propose that it was right in 2004 by David Lowe
Image scaling, rotation, affine transformation, illumination etc. have consistency and less sensitive to noise, and can complete characteristic point between image
Accurate matching.The basic ideas of SIFT detective operators are:Initially set up Gaussian difference scale space and from wherein detecting pole
It is worth point, then key point is accurately positioned and distributes principal direction, ultimately produces Feature Descriptor.It is seen that SIFT is a kind of non-
Often reliable characteristics of image point detecting method, it is easy to accomplish matching and positioning accuracy is higher, therefore it is quickly just in vision SLAM
Field is widely applied.
Prior art discloses a kind of fast image registration method based on local notable feature, its basic procedure such as Fig. 1
It is shown.
Mainly include the following steps:
A) reference picture and image to be matched being subjected to image enhancement, difference is down-sampled, in proportion downscaled images size, with
Reduce the feature extraction time;
B) obtained down-sampled image is subjected to SIFT feature extraction respectively;
C) use and characteristic point is clustered based on improved K-means clustering algorithms;
D) cluster subregion is carried out by cluster result, and cluster subregion is screened using image information entropy, cropped
Without matching area, the cluster subregion with correspondence is obtained;
E) carry out corresponding to the thick matching of characteristic point in cluster areas respectively, pick out remarkable characteristic;
F) marking area is selected centered on remarkable characteristic respectively in original reference image and image subject to registration;
G) SIFT essence matchings are carried out to obtained marking area.
Test result indicates that this method reducing the images match time, characteristic point quantity is controlled, is ensureing that matching is accurate
While spending, the efficiency of characteristic matching is effectively improved.
But the image registration algorithm based on local significantly SIFT feature has used improved K-means clustering algorithms to carry out
, it is necessary to determine clusters number in advance, larger cluster number can not only increase computation complexity, can also influence to match for cluster screening
Precision.
The content of the invention
A kind of the defects of present invention is directed to the prior art, there is provided SIFT feature based on the notable parameter index of regional area
Inspection optimization method, can effectively solve the problem that the above-mentioned problems of the prior art.
In order to realize above goal of the invention, the technical solution that the present invention takes is as follows:
A kind of SIFT map features point inspection optimization method based on the notable parameter index of regional area, its specific steps is such as
Under:
1) to current frame image, corresponding SIFT feature is detected first.
2) image is divided into M × N number of regional area, M represents line number, and N represents columns, and calculates each regional area
Characteristic remarkable parameter index value Ii, computational methods are as follows:
If j=1,2 ...;I=1,2 ..., M × N;
To containing njI-th of regional area of a SIFT feature, first obtains the average of wherein characteristic point position:If u generations
Table horizontal direction, v represent vertical direction;
Average on u direction:
Average on v directions:
And then obtain corresponding position variance:
J-th of characteristic point is in u direction upside deviation:
J-th of characteristic point is in v directions upside deviation:
Then have:
Its notable parameter index value of regional area without characteristic point is denoted as 0.
3) according to the I calculatediValue chooses K local marking area successively from high to low.
4) for each local marking area, m characteristic point is therefrom at most randomly selected.In order to obtain more preferable space point
Cloth, when selection, should ensure that the interval between this m SIFT feature is more than some pixel threshold.
5) selected SIFT feature is matched with current signature map.If successful match logarithm Match_num
Less than default threshold value Match_threshold, then need to just increase the number of characteristic point.At this moment, can be from remaining part
Marking area randomly selects (Match_threshold-Match_num) a SIFT feature, equally still ensures selected characteristic point
Between interval be more than default pixel threshold.Conversely, then without augmented feature point.
Compared with prior art the advantage of the invention is that:It is more advantageous to improving the convergence rate of monocular SLAM systems, together
When also contribute to more accurately describe scene environment.
Brief description of the drawings
Fig. 1 is fast image registration method flow chart of the prior art based on local notable feature;
Fig. 2 is the main flow chart of the embodiment of the present invention;
Fig. 3 is map feature point distribution map in prior art head two field pictures;
Fig. 4 is prior art map feature point and video camera relative position schematic diagram;
Fig. 5 is that map feature point is distributed in head two field pictures of the embodiment of the present invention;
Fig. 6 is the map feature point and video camera relative position schematic diagram of the embodiment of the present invention.
Embodiment
For the objects, technical solutions and advantages of the present invention are more clearly understood, develop simultaneously embodiment referring to the drawings, right
The present invention is described in further details.
The present invention from the statistical property of SIFT feature, fully ensure that be evenly distributed under the conditions of randomly select it is limited
The characteristic point of quantity, so as to fulfill both meeting that requirement of the EKF convergences to characteristic point quantity also ensure that the precision needed for scene description
It is required that.
The present invention proposes a kind of SIFT map features point inspection optimization method based on the notable parameter index of regional area, its
Comprise the following steps that:
1) to current frame image, corresponding SIFT feature is detected first.
2) image is divided into M × N number of regional area, and calculates the characteristic remarkable parameter index value of each regional area
Ii, computational methods are as follows:
To containing njI-th (i=1,2 ..., M × N) a regional area of (j=1,2 ...) a SIFT feature, is first asked
Go out the average of wherein characteristic point position:
Average on u direction:
Average on v directions:
And then obtain corresponding position variance:
J-th of characteristic point is in u direction upside deviation:
J-th of characteristic point is in v directions upside deviation:
Then have:
Note:Its notable parameter index value of regional area without characteristic point is denoted as 0.
3) according to the I calculatediValue chooses K local marking area successively from high to low.
4) for each local marking area, m characteristic point is therefrom at most randomly selected.In order to obtain more preferable space point
Cloth, when selection, should ensure that the interval between this m SIFT feature is more than some pixel threshold.
5) selected SIFT feature is matched with current signature map.If successful match logarithm Match_num
Less than default threshold value Match_threshold, then need to just increase the number of characteristic point.At this moment, can be from remaining part
Marking area randomly selects (Match_threshold-Match_num) a SIFT feature, equally still ensures selected characteristic point
Between interval be more than default pixel threshold, conversely, then without augmented feature point.
Pass through this above-mentioned 5 key steps, it becomes possible to obtain the limited scene map feature being evenly distributed in present frame
Point.The particular flow sheet of this method is as shown in Figure 2.
Here is the simulating, verifying of the method for the present invention:
1) experimental situation and platform introduction
Using the OEM LD042 monocular cams by calibration, its resolution ratio is 320 × 240, and output image form is
.JPG, frame speed is 20fps.The configuration of hardware and software platform is shown in Table 1.
The hardware and software platform of 1 emulation experiment of table
Experimental situation, which is chosen, tests indoor somewhere scene, and ensures that illumination condition does not have violent change.Tested
When, hand-held monocular cam is slowly smoothly moved in a small range, and is continuously shot 20s or so.Then by the multiframe of acquisition
Image sequence carries out the off-line simulation experiment of SIFT map features point detection as input.
2) simulation parameter is set
In order to estimate the position of the movement locus of video camera and surrounding scene road sign characteristic point by emulation experiment, now set
It is as follows to put main simulation parameter:
Video camera starting position coordinates:(0 0 0)T
The original state of 4 element representations:(1 0 0 0)T
Camera intrinsic parameter matrix:
Controlling cycle:DT_CONTROL=0.05s
Velocity noise maximum:SigmaV=0.03m/s
Angle noise maximum:SigmaG=(3.0*pi180)
Control the covariance of noise:
Observation cycle:DT_OBSERVE=0.4s
Observe ultimate range:R_MAX=2.5m
Observed range noise maximum:SigmaD=0.02m
Observation angle noise maximum:SigmaA=(2.0*pi180)
The covariance of observation noise:
3) simulation result and analysis
When carrying out off-line simulation experiment, when directly using SIFT detective operators, the characteristic point in first two field picture is obtained
Initial distribution situation is as shown in Figure 3,4.
121 stable SIFT features are extracted altogether from the point of view of the result that Fig. 4 is shown, they are distributed in image mostly
The larger boundary of grey value difference.
When the detection scheme proposed in using the present invention, the parameter in experiment is chosen for:M × N=8 × 6, K=30, m
=2, Match_threshold=30, and the spacing distance for randomly selecting characteristic point is not less than 15 pixels.Based on improved
SIFT feature distribution situation is as shown in Figure 5,6 in the first two field picture of SIFT feature extraction algorithm.
Note:Whole image region is divided into 8 × 6 square regions by blue line in figure;Red is marked as by each partial zones
The sequence number that the notable parameter index in domain arranges from high to low;Green '+' point represents the map obtained according to improvement SIFT extracting methods
The position of characteristic point.
After improvement, the map feature quantity for making extraction by arrange parameter is maintained at 60 or so, and is shown according to regional area
Preferably, mutual spacing is obvious for the characteristic point dispersiveness that work parameter index is selected.
Those of ordinary skill in the art will understand that the embodiments described herein, which is to help reader, understands this hair
Bright implementation, it should be understood that protection scope of the present invention is not limited to such special statement and embodiment.Ability
The those of ordinary skill in domain can according to the present invention these disclosed technical inspirations make it is various do not depart from essence of the invention its
Its various specific deformations and combination, these deformations and combination are still within the scope of the present invention.
Claims (1)
- A kind of 1. SIFT map features point inspection optimization method based on the notable parameter index of regional area, it is characterised in that specific Step is as follows:1) to current frame image, corresponding SIFT feature is detected first;2) image is divided into M × N number of regional area, M represents line number, and N represents columns, and calculates the feature of each regional area Notable parameter index value Ii, computational methods are as follows:If j=1,2 ...;I=1,2 ..., M × N;To containing njI-th of regional area of a SIFT feature, first obtains the average of wherein characteristic point position:If u represents level Direction, v represent vertical direction;Average on u direction:Average on v directions:And then obtain corresponding position variance:J-th of characteristic point is in u direction upside deviation:J-th of characteristic point is in v directions upside deviation:Then have:Its notable parameter index value of regional area without characteristic point is denoted as 0;3) according to the I calculatediValue chooses K local marking area successively from high to low;4) for each local marking area, no more than m characteristic point is therefrom randomly selected;In order to obtain more preferable space point Cloth, when selection, should ensure that the interval between this m SIFT feature is more than some pixel threshold;5) selected SIFT feature is matched with current signature map, if successful match logarithm Match_num is less than Default threshold value Match_threshold, then need to just increase the number of characteristic point;At this moment, from remaining local marking area (Match_threshold-Match_num) a SIFT feature is randomly selected, between equally still ensureing between selected characteristic point Every more than default pixel threshold, conversely, then without augmented feature point.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711264919.1A CN108022263A (en) | 2017-12-05 | 2017-12-05 | A kind of SIFT feature inspection optimization method based on the notable parameter index of regional area |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711264919.1A CN108022263A (en) | 2017-12-05 | 2017-12-05 | A kind of SIFT feature inspection optimization method based on the notable parameter index of regional area |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108022263A true CN108022263A (en) | 2018-05-11 |
Family
ID=62078554
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711264919.1A Pending CN108022263A (en) | 2017-12-05 | 2017-12-05 | A kind of SIFT feature inspection optimization method based on the notable parameter index of regional area |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108022263A (en) |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103988226A (en) * | 2011-08-31 | 2014-08-13 | Metaio有限公司 | Method for estimating camera motion and for determining three-dimensional model of real environment |
CN104794210A (en) * | 2015-04-23 | 2015-07-22 | 山东工商学院 | Image retrieval method combining visual saliency and phrases |
-
2017
- 2017-12-05 CN CN201711264919.1A patent/CN108022263A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103988226A (en) * | 2011-08-31 | 2014-08-13 | Metaio有限公司 | Method for estimating camera motion and for determining three-dimensional model of real environment |
CN104794210A (en) * | 2015-04-23 | 2015-07-22 | 山东工商学院 | Image retrieval method combining visual saliency and phrases |
Non-Patent Citations (2)
Title |
---|
李文凤: "基于图像显著区域检测的SIFT特征匹配方法研究", 《微型机与应用》 * |
胡衡: "基于单目视觉的SLAM方法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Zhang et al. | Active exposure control for robust visual odometry in HDR environments | |
CN105631880B (en) | Lane line dividing method and device | |
CN105488811B (en) | A kind of method for tracking target and system based on concentration gradient | |
CN105046206B (en) | Based on the pedestrian detection method and device for moving prior information in video | |
Alt et al. | Rapid selection of reliable templates for visual tracking | |
CN104835175B (en) | Object detection method in a kind of nuclear environment of view-based access control model attention mechanism | |
CN105930822A (en) | Human face snapshot method and system | |
CN108198201A (en) | A kind of multi-object tracking method, terminal device and storage medium | |
CN103729649B (en) | A kind of image rotation angle detection method and device | |
CN104517102A (en) | Method and system for detecting classroom attention of student | |
CN107240112B (en) | Individual X corner extraction method in complex scene | |
CN107424142A (en) | A kind of weld joint recognition method based on saliency detection | |
CN104008542B (en) | A kind of Fast Corner matching process for specific plane figure | |
CN107993258A (en) | A kind of method for registering images and device | |
CN107066969A (en) | A kind of face identification method | |
CN106355607B (en) | A kind of width baseline color image template matching method | |
CN108171715A (en) | A kind of image partition method and device | |
Venkatesan et al. | Face recognition system with genetic algorithm and ANT colony optimization | |
JP2009163682A (en) | Image discrimination device and program | |
CN103733225B (en) | Characteristic point peer system, characteristic point counterpart method and record medium | |
CN110348366B (en) | Automatic optimal face searching method and device | |
CN109063598A (en) | Face pore detection method, device, computer equipment and storage medium | |
CN108491857A (en) | A kind of multiple-camera target matching method of ken overlapping | |
CN106897730A (en) | SAR target model recognition methods based on fusion classification information with locality preserving projections | |
CN105631849B (en) | The change detecting method and device of target polygon |
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: 20180511 |
|
RJ01 | Rejection of invention patent application after publication |