CN113313659A - High-precision image splicing method under multi-machine cooperative constraint - Google Patents
High-precision image splicing method under multi-machine cooperative constraint Download PDFInfo
- Publication number
- CN113313659A CN113313659A CN202110450860.5A CN202110450860A CN113313659A CN 113313659 A CN113313659 A CN 113313659A CN 202110450860 A CN202110450860 A CN 202110450860A CN 113313659 A CN113313659 A CN 113313659A
- Authority
- CN
- China
- Prior art keywords
- image
- point
- splicing
- points
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 58
- 239000011159 matrix material Substances 0.000 claims abstract description 28
- 238000009499 grossing Methods 0.000 claims abstract description 6
- 230000009466 transformation Effects 0.000 claims description 23
- 230000003044 adaptive effect Effects 0.000 claims description 16
- 238000012937 correction Methods 0.000 claims description 15
- 238000003384 imaging method Methods 0.000 claims description 15
- 238000013507 mapping Methods 0.000 claims description 6
- 238000000354 decomposition reaction Methods 0.000 claims description 5
- 238000005070 sampling Methods 0.000 claims description 4
- 238000012216 screening Methods 0.000 claims description 4
- RZVHIXYEVGDQDX-UHFFFAOYSA-N 9,10-anthraquinone Chemical compound C1=CC=C2C(=O)C3=CC=CC=C3C(=O)C2=C1 RZVHIXYEVGDQDX-UHFFFAOYSA-N 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000005259 measurement Methods 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 claims description 3
- 239000000126 substance Substances 0.000 claims description 3
- 230000008569 process Effects 0.000 abstract description 8
- 230000000694 effects Effects 0.000 abstract description 5
- 238000007689 inspection Methods 0.000 abstract description 3
- 230000007704 transition Effects 0.000 abstract description 3
- 238000012795 verification Methods 0.000 abstract description 2
- 238000011156 evaluation Methods 0.000 description 9
- 230000000007 visual effect Effects 0.000 description 9
- 238000002474 experimental method Methods 0.000 description 7
- 238000001514 detection method Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 238000005457 optimization Methods 0.000 description 4
- 238000000605 extraction Methods 0.000 description 3
- 230000036544 posture Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000000746 purification Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 241000422846 Sequoiadendron giganteum Species 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 239000004576 sand Substances 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a high-precision image stitching method under multi-machine cooperative constraint, which starts from the multi-machine cooperative control angle and builds a multi-scale multi-view image stitching model based on the space pose information of multiple unmanned aerial vehicles. Aiming at the phenomena of double images, pixel breakage and the like of a single homography matrix, the self-adaptive homography matrix method is provided for improving the splicing effect. For the problems of fuzzy identical overlapping areas and color inconsistency after splicing, a weighted smoothing algorithm is adopted, and smooth transition of overlapping partial images is realized through weight distribution, so that the problem of color difference near splicing superposition is effectively solved. The self-contained aerial photography data set is used for algorithm performance verification, and experimental results show that the method provided by the invention has good splicing performance, obviously improves the registration precision, and meets the requirements of high real-time performance and high precision for splicing the multi-scale and multi-view aerial photography images in the multi-machine collaborative inspection flight process.
Description
Technical Field
The invention relates to the technical field of image data processing, in particular to a high-precision image splicing method under multi-machine cooperative constraint.
Background
Image stitching is a process of combining a plurality of images with overlapping regions in the same scene into a complete image with wide visual field and high resolution. Currently common image registration methods include image gray scale based, transform domain based, and feature based methods. The Feature-based matching method is widely applied, and includes basic HOG (Histogram of Oriented Gradients) Features, Scale-Invariant Feature Transform (SIFT) algorithm, speedup Robust Features (SURF) algorithm, and ORB (Oriented FAST and Robust BRIEF) and other improved algorithms. The HOG characteristics have invariance and strong robustness to image geometry and optical deformation, but have poor real-time performance, sensitivity to noise points and difficult processing to the shielding problem; the SIFT algorithm has strong robustness and high reliability, but the calculated amount is large, and the requirement of splicing instantaneity cannot be met; the SURF algorithm has high matching speed and high operation efficiency, but is not ideal in performance such as scale invariance, rotation invariance and the like; the ORB algorithm has high calculation speed, the running time of the ORB algorithm is far superior to SIFT and SURF, the real-time performance is good, and the ORB algorithm has invariance to noise and perspective transformation thereof. In addition, patent CN105869120A discloses a real-time optimization method for image stitching, which reduces the search range of feature points by locking the overlapping area, and increases the stitching speed, and the method focuses on high real-time performance of stitching, but the stitching quality is not high. Patent CN108921780A discloses a method for quickly stitching sequence images of unmanned aerial vehicle aerial photography, which has good stability and low redundant information, but image detail information is easy to lose, and is not beneficial to generating a high-precision stitching result image.
Because the camera visual angle is limited and the environment is complicated, be difficult to obtain high resolution, comprehensive wide area result picture in real time, and utilize many unmanned aerial vehicles to carry on multimode sensor and patrol the mission zone in coordination and can improve and patrol efficiency by a wide margin, nevertheless still have two problems when using unmanned aerial vehicle to patrol: the information sharing and communication problems among multiple machines are solved, the transmission of complex mass data is completed in a very short time, and the information asynchronization is avoided; secondly, the post-processing problem of images shot by multiple machines is solved, the image information acquired by the multiple machines is separated, and the aerial images acquired by the sensors have different scales and visual angles due to the fact that the flying heights, relative positions and postures of the unmanned aerial vehicles are different. At present, the splicing effect of the unmanned aerial vehicle aerial image splicing method on the same plane image is ideal, but the conditions of mutual intersection, overlapping and the like of a plurality of scales and a plurality of visual angle images often occur in multi-machine collaborative shooting, and the existing algorithm cannot meet the requirements of high real-time performance and high precision at the same time.
In order to solve the problem, how to reasonably utilize a multi-machine cooperation strategy to realize the splicing of multi-scale and multi-view aerial images becomes a problem to be solved urgently at present.
Disclosure of Invention
Aiming at the existing problems, the invention provides a high-precision image splicing method under multi-machine cooperative constraint, which can solve the problems of low splicing efficiency and poor quality caused by large data volume and complex types in the cooperative inspection process of multiple unmanned aerial vehicles.
The core thought of the invention is as follows:
firstly, from the perspective of multi-machine cooperative control, acquiring aerial images in real time, and selecting a rapid splicing scheme of splicing among unmanned aerial vehicles and splicing among sequence frames of single machines; then, an ORB feature extraction algorithm is utilized, the difference of image transformation relations of different scales and different visual angles under the condition of multi-camera shooting is fully considered, and a self-adaptive homography matrix is constructed to complete high-precision registration and splicing of the images; and finally, realizing smooth transition of the overlapped images by using a weighted smoothing algorithm, and solving the problems of visual effects such as fracture, obvious seam and the like of the spliced large-view-field image, thereby improving the image splicing quality.
The technical solution for realizing the purpose of the invention is as follows:
a high-precision image splicing method under multi-machine cooperative constraint is characterized by comprising the following steps:
step 1: acquiring ground images in real time through multiple sensors, and constructing a multi-machine cooperation real-time image imaging model by combining position, posture and angle information among multiple unmanned aerial vehicles, so as to determine an actual imaging area;
step 2: extracting a characteristic point set of an original image in an actual imaging area by adopting an ORB (object-oriented bounding box) method, performing image rough matching, purifying a rough matching result by using a random sampling consistency algorithm, and removing mismatching points;
and step 3: constructing an adaptive homography matrix, mapping the purified images to the same coordinate system for preliminary splicing, then performing time sequence correction on each sequence of images, and then performing color correction by adopting a weighted smoothing algorithm;
and 4, step 4: and outputting a result image after finishing image splicing.
Further, the specific operation steps of step 1 include:
step 11: resolving the acquired ground image data to obtain the motion parameters of the unmanned aerial vehicle, and realizing the flight control of the airframe;
step 12: utilize inertial navigation unit, GPS and the barometer of carrying on many unmanned aerial vehicles to obtain unmanned aerial vehicle attitude angle: pitch angle theta, roll angle phi, yaw angleAnd information such as coordinates, flying height, etc.;
step 13: respectively establishing a ground coordinate system, a body coordinate system, a camera coordinate system and an image coordinate system;
step 14: coordinate transformation relation p between body coordinate system and ground coordinate system is established by combining attitude angle of unmanned aerial vehicleg=LbpbAnd establishing a multi-machine cooperation real-time image imaging model.
Further, the specific operation steps of step 2 include:
step (ii) of21: selecting any pixel point S in the original image, drawing a circle with the radius of 3 pixels by taking S as the center of the circle, detecting 16 pixel points falling on the circle, recording the number of continuous pixel points with the pixel gray scale satisfying the formula (2) in the 16 pixel points on the neighborhood circle as h, and judging whether h is greater than a preset threshold value epsilondIf yes, judging S as a characteristic point, and the gray value condition satisfied by the pixel is as follows:
wherein I (x) is the gray scale of any point on the circumference, I(s) is the gray scale of the center of the circle, epsilondIs a threshold value of the gray scale difference, and N represents the gray scale difference;
step 22: directly detecting the gray values of four pixel positions of each pixel point S in the circumferential vertical direction I (1) and I (9) and the horizontal direction I (5) and I (13), and counting that the difference value of the gray values of the four position points I (t) and the pixel of the selected point is greater than epsilondThe number of the pixel points M is as follows:
M=size(I∈{|I(t)-I(s)|>εd}),εd>0 (3);
if the characteristic point satisfies the formula (3) and M33, judging S as a characteristic point, otherwise, directly excluding the point;
step 23: calculating Hamming distances d between any one feature point S selected from one image and all feature points V in other images, sequencing the obtained distances d, selecting a point with the closest distance as a matching point, and establishing a rough matching point pair to form a feature point set;
step 24: and screening the obtained feature point set by adopting a random sampling consistency method, and finally removing mismatching points to obtain a purified matching feature point set.
Further, the specific operation steps of step 24 include:
step 241: selecting Q points on the obtained characteristic point set, and calculating all characteristic points P of the first image according to the set registration straight line model1Set of mapped points P on the second image* 2And satisfy the mappingThe relationship is as follows:and m is 1,2, …, Q;
step 242: computingEach point and the second image feature point set P2Euclidean distance of the corresponding point, setting the threshold value of the gray level difference of the characteristic points as delta, and counting P* 2The number N of the characteristic points satisfying that the Euclidean distance is less than deltaT:
Step 243: re-randomly selecting Q points, repeating the K steps 241-242, and recording N timesTUntil the iteration is finished;
step 244: selecting N as max (N)Tk(K-1, 2, …, K)) registration model as the final fitting result, thus rejecting the mis-match points.
Further, the specific operation steps of step 3 include:
step 31: adopting a mobile DLT method to realize linear fitting to the self-adaptive homography matrix according to the coordinate values of the plurality of groups of purified matching target point pairs;
step 32: time sequence correction is carried out on the multi-unmanned aerial vehicle aerial image sequence by adopting the image central point position after homography transformation, and time sequence moving parameter flag of the corresponding imageoThe values of (A) are as follows:
wherein, PoxIs the x coordinate of the central point of the o-th homography transformed image, and L is the width of the reference image;
step 33: traversing each frame of the image sequence according to flagoThe different values of (2) are corrected: when flagoWhen the number is 1, the next frame of the image to be spliced is used forSplicing, when flagoWhen the image number is-1, the previous frame of the image to be spliced is used for splicing until a flagoWhen the value is 0, outputting the corrected image sequence;
step 34: and recalculating new pixel values of the overlapped area by using the linear gradient of the image, and performing color correction.
Further, the specific operation steps of step 31 include:
step 311: the feature matching point sets of the two images are respectively X*' and X*Constructing an adaptive homography transformation relation X*'=H*X*;
Step 312: changing the constructed adaptive homography conversion relation into an implicit form, and linearizing into: x*'×H*X*When the value is equal to 0, the following can be obtained:
step 313: memory linearization matrixCan obtain A h 0, for A*Singular value decomposition is carried out to obtain an adaptive homography matrix H*Right singular vector of
Wherein, the weight valueIndicates a control point x*The more recent pixel data is important, the variance σ2Is a measure of how far the measurement point deviates from the control point;
step 314: according to the obtained right singular vectorI.e. a reconstructable adaptive homography matrix H*。
Further, the formula for calculating the new pixel value of the overlap region in step 34 is:
I(x,y)=Il(xl,yl)(1-α)+Ir(xr,yr)α (6),
wherein the content of the first and second substances,represents the distance from the abscissa of the pixel point to the abscissa of the left boundary of the overlap region, (x)l,yl) To stitch the front left image coordinates, (x)r,yr) For stitching front-right image coordinates, Il(xl,yl) For stitching front left image pixels, Ir(xr,yr) To stitch the front right image pixels, (x)lmax,ylmax) The maximum value of the left image coordinates before splicing, (x, y) are the image coordinates after splicing, and I (x, y) is the image pixel after splicing.
Compared with the prior art, the method has the following beneficial effects:
firstly, aiming at the transformation problem of images with different scales and different visual angles under the complex and changeable environment and the cooperation of multiple machines, the invention provides a self-adaptive homography matrix method, which basically eliminates the phenomena of double images, pixel fracture and the like existing in the traditional homography matrix.
Secondly, the invention also adopts a weighted smoothing algorithm, realizes the smooth transition of the overlapped part images through weight distribution, and effectively solves the problems of fuzzy overlapped areas and inconsistent colors after splicing by combining a linear gradual change thought.
In conclusion, the registration precision of the characteristics of the mosaic obtained by the method provided by the invention is obviously improved, and the high-precision requirement of mosaic is met.
Drawings
FIG. 1 is a flow chart of a stitching algorithm of the present invention;
FIGS. 2(a) - (c) are schematic diagrams of the transformation of each coordinate system;
FIG. 3 is a schematic diagram of feature point matching of four scene aerial photography source images;
FIG. 4 is two different sets of aerial source images;
FIG. 5 is a splicing experimental graph of two groups of different aerial photography source images;
fig. 6 is a multi-scale multi-view image stitching experimental diagram under four scenes.
Detailed Description
In order to make those skilled in the art better understand the technical solution of the present invention, the following further describes the technical solution of the present invention with reference to the drawings and the embodiments.
The invention provides a high-precision image splicing optimization method under multi-machine cooperative constraint, which comprises the following steps of:
step 1: acquiring images in real time by using multiple sensors, and establishing a multi-machine cooperation real-time image imaging model by combining position, posture and angle information among multiple unmanned aerial vehicles;
the multi-sensor real-time acquisition ground image, solve image data and handle and obtain unmanned aerial vehicle motion parameter, realize the flight control of organism. The attitude angles (pitch angle theta, roll angle phi and yaw angle) of the unmanned aerial vehicles are obtained by combining inertial navigation units (IMUs), GPS and barometers carried on the unmanned aerial vehicles) Coordinates and fly height. A multi-machine collaborative imaging model is established based on pose information acquired by each module carried by the unmanned aerial vehicle, so that feature information of a patrol area can be conveniently acquired. The method comprises the following specific steps:
(1) establishing a coordinate system
First, a ground coordinate system p is establishedg=(xg,yg,zg)TAssuming that the ground is a plane, the points on the ground and the obstacle are all located on the plane. Wherein, the coordinate (x)g,yg,zg) For ground points in the x, y, z directionsThe component of (a), pointing to satisfy the right-hand rule of the coordinate system;
secondly, a coordinate system P of the body is establishedb=(xb,yb,zb)TSelecting the center of mass of the body as the origin, wherein xbAxis directed to the head, ybAxis pointing to the right of the fuselage, zbThe shaft points to the lower part of the machine body;
thirdly, a camera coordinate system p is establishedci=(xci,yci,zci)TAssuming that each drone i corresponds to a camera, the cameras can be placed in any position, and therefore the ground coordinates are chosen to describe the camera position. The relation between the camera coordinate system and the ground coordinate system can be determined by a rotation matrix RciAnd a translation vector tciIs described as pg=Rcipci+tci(ii) a The transformation relationship between the ground coordinate system and the camera coordinate system is shown in figure 2(a),
the conversion relationship between the body coordinate system and the ground coordinate system is shown in fig. 2 (b);
finally, an image coordinate system p is establishedIi=(xI,yI)TThe image coordinate system is a two-dimensional coordinate system established for the real-time image, the upper left corner of the real-time image is taken as the origin of coordinates, and xIAxis horizontal to right, yIAxis perpendicular xIAxially downwards, the coordinate transformation relation between the image coordinate system and the camera coordinate system is as follows: p is a radical ofIi=RIipciWherein the matrix R is transformedIiThe transformation of the image coordinate system to the camera coordinate system, in relation to the camera parameters, is shown in fig. 2 (c).
(2) Establishing an imaging model
When the airborne camera is used for imaging, the imaging area can be regarded as intersection of a rectangular pyramid and a plane, and the actual imaging area can be determined through linear equations of four edges of each unmanned aerial vehicle. And establishing a coordinate transformation relation between a body coordinate system and a ground coordinate system by combining the attitude angle of the unmanned aerial vehicle: p is a radical ofg=Lbpb. Transformation matrix L around three axes x, y, zbThe attitude angles theta, phi and phi with the unmanned aerial vehicle,In this regard, it can be expressed as:
if there are n image coordinates p that unmanned aerial vehicle shotIi(i ═ 1,2, L, n) and pgCorrespond to, i.e. pgIn the imaging range of n unmanned aerial vehicles, the image fusion can be carried out on the part of images in the later stage.
Step 2: and (3) extracting gray value information of the continuous pixel points by adopting an ORB method, further judging image characteristic points, and performing characteristic matching and purification on the extracted characteristic points. The method comprises the following specific steps:
(1) feature extraction
The method comprises the steps of adopting an ORB method, judging image feature points based on gray value information of continuous pixel points, comparing the gray value size result of any pixel point S in an image with the gray value size result of a point in a circular neighborhood, and extracting a source image feature point set, wherein the specific process is described as follows:
firstly, global detection is carried out on candidate points: selecting any pixel point S in the image, drawing a circle with the radius of 3 pixels by taking S as the center of the circle, considering 16 pixels on the circumference of the pixel point, detecting the 16 pixel points falling on the circumference, recording the number of continuous pixel points with the gray scale of the pixels in the 16 pixel points on the neighborhood circle satisfying the formula (2) as h, and judging whether h is greater than a preset threshold value epsilondNormally, the threshold is set to 12, i.e., if h>If the gray value of the pixel meets the condition that S is judged as the characteristic point and the gray value corresponding to the pixel meets the condition that:
wherein I (x) is the gray scale of any point on the circumference, I(s) is the gray scale of the center of the circle, epsilondN represents a gray difference value as a threshold value of the gray difference value.
Secondly, carrying out optimization detection on the candidate points: the ORB optimization detection method is utilized to accelerate the extraction of the characteristic points,and the detection efficiency is improved. Directly detecting the gray values of four pixel positions of each pixel point S in the circumferential vertical direction I (1) and I (9) and the horizontal direction I (5) and I (13), and counting the gray difference value of the four position points I (t) (t takes 1, 5, 9 and 13 respectively) and the pixel gray value of the selected point is more than epsilondThe number of the pixel points M is as follows:
M=size(I∈{|I(t)-I(s)|>εd}),εd>0 (3),
if S satisfies formula (3) and M3And 3, judging S as a characteristic point, and if not, directly excluding the point. Therefore, the feature point set of each image in the original image is screened out, and subsequent feature matching is facilitated.
(2) Feature matching and purification
Firstly, carrying out feature matching on the extracted feature points: and calculating the Hamming distance d of the features, and taking the Hamming distance d as an evaluation criterion of feature matching similarity. Measuring Hamming distance between any characteristic point S in image and all characteristic points V in other imagesWherein i*0,1, …, n-1, denotes an encoding where S and V are both n bits,is an exclusive or operation. And sequencing the obtained distances, selecting one closest to the distances as a matching point, and establishing a rough matching point pair. And a large number of mismatching exists in the matching result, and mismatching points need to be filtered by means of a screening mechanism.
Secondly, feature purification is carried out on the matched feature point set: screening the feature point set by using a Random Sample Consensus (RANSAC) method, selecting Q points, and calculating all feature points P of the first image according to an assumed registration straight line model1Set of mapped points P on the second image* 2And satisfy the mapping relationWherein m is 1,2, …, Q;
computingEach point and the second image feature point set P2Setting the Euclidean distance of the corresponding points, setting the threshold value of the gray level difference of the characteristic points as delta, and counting P* 2The number N of the characteristic points satisfying that the Euclidean distance is less than deltaT:
Re-randomly selecting Q points, repeating the above operation K times and recording N timesTUntil the iteration is finished;
finally, selecting N as max (N)Tk(K-1, 2, …, K)) registration model as the final fitting result, rejecting mis-matching points.
And step 3: and constructing an adaptive homography matrix, mapping the images to the same coordinate system for splicing, performing time sequence correction on each sequence of images, and performing color correction by adopting a weighted smoothing algorithm to finish image splicing. The method comprises the following specific steps:
(1) adaptive homography transformation
Calculating homography matrixes among aerial images, mapping the images to the same coordinate system according to the homography matrixes, and splicing the images into a rough image;
and then using a moving DLT (Direct Linear transform) method to realize linear fitting of the homography matrix according to the matched target point, wherein the fitting process is as follows: assuming that the feature matching point sets of the two images are X and X ', constructing a homography transformation relation X' ═ HX, namely:
(x' y' 1)T=H(x y 1)T,
wherein, (x 'y' 1)TAs coordinates of the feature point in X' (X y 1)TH is a homography matrix of 3 × 3, which is the coordinates of the feature points in X, i.e.:
it is changed to implicit form and linearized as: x' X HX ═ 0, then:
memory linearization matrixThen, Ah is obtained as 0 and h is solved. Take the first two rows a of AsAnd maps it to A ∈ R2N*9SVD (singular value Decomposition) Decomposition is performed on A, anddefining right singular vector corresponding to minimum singular value of A, obtainingH can be formed byReconstructing the mixture;
finally, based on the construction process of the homography matrix H, the position-dependent self-adaptive homography matrix H is designed*Satisfy the homography transformation relation X*'=H*X*And the weight is given to adapt to the transformation process of the data, so that the limitation that the traditional single homography matrix is only applicable in the same plane aiming at the view angle range is avoided, and the problem of ghosting or registration error caused by single homography transformation is solved. The method comprises the following specific operation steps:
step 1: the feature matching point sets of the two images are respectively X*' and X*Constructing an adaptive homography transformation relation X*'=H*X*;
Step 2: transforming the constructed adaptive homography into a relational expressionIn implicit form, and is linearized as X*'×H*X*When the value is equal to 0, the following can be obtained:
and step 3: memory linearization matrixCan obtain A h 0, for A*Singular value decomposition is carried out to obtain an adaptive homography matrix H*Right singular vector of
Wherein, the weight valueIndicates a control point x*The more recent pixel data is important, the variance σ2Is a measure of how far the measurement point deviates from the control point;
and 4, step 4: obtained according to the formula (4)And reconstructObtaining an adaptive homography matrix H*。
(2) Timing correction
The invention adopts the image central point position after homography transformation to carry out time sequence correction on the aerial image sequence of the multiple unmanned aerial vehicles, and the correction formula is as follows:
wherein, PoxIs the x coordinate of the central point of the o-th homography transformed image, L is the width of the reference image, and flagoIs a time sequence moving parameter of a corresponding image, and the value is as follows:
traversing each frame of the image sequence until a flagoAnd outputting the corrected image sequence when the image sequence is equal to 0.
(3) Color correction
Aiming at the problems of blurring, ghosting and color inconsistency of the same overlapped area after splicing, the linear gradual change of the image is adopted to recalculate a new pixel value of the overlapped area, and the splicing effect is optimized through color correction.
Assuming that alpha represents the distance from the abscissa of the pixel point to the abscissa of the left boundary of the overlapping region, the gray value of the pixel point of the spliced image is given by the following formula:
I(x,y)=Il(xl,yl)(1-α)+Ir(xr,yr)α (6),
wherein the content of the first and second substances,(xl,yl) To stitch the front left image coordinates, (x)r,yr) For stitching front-right image coordinates, Il(xl,yl) For stitching front left image pixels, Ir(xr,yr) To stitch the front right image pixels, (x)lmax,ylmax) The maximum value of the left image coordinates before splicing, (x, y) are the image coordinates after splicing, and I (x, y) is the image pixel after splicing.
By the processing, a good splicing effect can be realized, and a well spliced image is obtained.
Examples
The verification of the invention is completed in a multi-unmanned aerial vehicle cooperative target detection and identification system with the registration number of 2020SR 1088587.
1. Experimental Environment
CPU:Intel Xeon E5-1650 v4;
RAM:32GB;
NVIDIA TITAN-X, Windows10 system, Visual studio 2015+ Anaconda3.5+ Python 3.6.
2. Procedure of experiment
The experiment utilizes the aerial image that many unmanned aerial vehicle flight platforms gathered in the tour flight experiment, verifies the performance of the concatenation algorithm that provides, specifically includes two experiments:
(1) splicing experiment for two groups of different aerial photography source images
According to the steps of the splicing method provided by the invention, two groups of aerial source images from different sources are effectively spliced as shown in figure 4, the splicing position is smooth, and the splicing result is shown in figure 5. As can be seen from the figure, the method effectively expands the visual angle range of the image, retains the respective image characteristics of the two source images, and effectively displays the key information of pits, people, lamp boards and the like which are shielded by the big tree in the source images.
(2) Multi-scale and multi-view image splicing experiment under four scenes
Four experimental scenes are constructed by using a plurality of unmanned aerial vehicles to patrol scenes of key roads in a complex area, the matching schematic diagram of feature points of the aerial source images in the four scenes is shown in figure 3, the image splicing is carried out in each scene according to each step of the splicing method provided by the invention, and the splicing result is shown in figure 6. As can be seen from the figure, the method has high splicing precision and no image distortion.
3. Evaluation of splicing results
In order to fully verify the splicing method provided by the invention, the splicing effect is objectively and comprehensively evaluated by adopting evaluation parameters such as RMSE (Root Mean Square Error), Dice distance, Hausdorff distance and the like.
(1) RMSE evaluation
The Root Mean Square Error (RMSE) evaluation refers to the square root of the mean deviation between a plurality of characteristic points of the image to be registered and the pixel point position of the transformed reference image after the two images are registered:
wherein, Δ xqAnd Δ yqThe error in the x-direction and y-direction for the q-th registered feature point.
(2) Dice distance evaluation
The Dice distance is used to represent the degree to which two sets contain each other:
wherein, TP represents the logarithm of correct matching points, GT and PR represent the feature point sets of the target image and the transformation image respectively.
(3) Hausdorff distance evaluation
The Hausdorff distance is used to calculate the match of the two sets of points:
H(A,B)=max(h(A,B),h(B,A)),
wherein h (A, B) is maxa∈Aminb∈B||a-b||2,h(B,A)=maxb∈Bmina∈A||b-a||2。
According to the three evaluation indexes, the method provided by the invention and the registration algorithm based on SURF and SIFT images are compared and analyzed, and the obtained comparison result is shown in Table 1:
TABLE 1 comparison of quantization indices for different methods
As can be seen from the table 1, the method provided by the invention is superior to SURF and SIFT algorithms on three evaluation criteria of RMSE, Dice and Hausdorff, the running time of the algorithm is short, the matching error of the feature points is obviously reduced, the complex and changeable environment of the algorithm experiment is met, and the requirements of high instantaneity and high precision of splicing the multi-scale and multi-view aerial images in the multi-machine collaborative inspection flight process are met.
Those not described in detail in this specification are within the skill of the art. Although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that various changes in the embodiments and modifications of the invention can be made, and equivalents of some features of the invention can be substituted, and any changes, equivalents, improvements and the like, which fall within the spirit and principle of the invention, are intended to be included within the scope of the invention.
Claims (7)
1. A high-precision image splicing method under multi-machine cooperative constraint is characterized by comprising the following steps:
step 1: acquiring ground images in real time through multiple sensors, and constructing a multi-machine cooperation real-time image imaging model by combining position, posture and angle information among multiple unmanned aerial vehicles, so as to determine an actual imaging area;
step 2: extracting a characteristic point set of an original image in an actual imaging area by adopting an ORB (object-oriented bounding box) method, performing image rough matching, purifying a rough matching result by using a random sampling consistency algorithm, and removing mismatching points;
and step 3: constructing an adaptive homography matrix, mapping the purified images to the same coordinate system for preliminary splicing, then performing time sequence correction on each sequence of images, and then performing color correction by adopting a weighted smoothing algorithm;
and 4, step 4: and outputting a result image after finishing image splicing.
2. The method for splicing high-precision images under multi-machine cooperative constraint according to claim 1, wherein the specific operation steps of step 1 comprise:
step 11: resolving the acquired ground image data to obtain the motion parameters of the unmanned aerial vehicle, and realizing the flight control of the airframe;
step 12: utilize inertial navigation unit, GPS and the barometer of carrying on many unmanned aerial vehicles to obtain unmanned aerial vehicle attitude angle: pitch angle theta, roll angle phi, yaw angleAnd information such as coordinates, flying height, etc.;
step 13: respectively establishing a ground coordinate system, a body coordinate system, a camera coordinate system and an image coordinate system;
step 14: coordinate transformation relation p between body coordinate system and ground coordinate system is established by combining attitude angle of unmanned aerial vehicleg=LbpbAnd establishing a multi-machine cooperation real-time image imaging model.
3. The method for splicing images with high precision under the multi-machine cooperative constraint as claimed in claim 1, wherein the specific operation steps of step 2 include:
step 21: selecting any pixel point S in the original image, drawing a circle with the radius of 3 pixels by taking S as the center of the circle, detecting 16 pixel points falling on the circle, recording the number of continuous pixel points with the pixel gray scale satisfying the formula (2) in the 16 pixel points on the neighborhood circle as h, and judging whether h is greater than a preset threshold value epsilondIf yes, judging S as a characteristic point, and the gray value condition satisfied by the pixel is as follows:
wherein I (x) is the gray scale of any point on the circumference, I(s) is the gray scale of the center of the circle, epsilondIs a threshold value of the gray scale difference, and N represents the gray scale difference;
step 22: directly detecting the gray values of four pixel positions of each pixel point S in the circumferential vertical direction I (1) and I (9) and the horizontal direction I (5) and I (13), and counting that the difference value of the gray values of the four position points I (t) and the pixel of the selected point is greater than epsilondThe number of the pixel points M is as follows:
M=size(I∈{|I(t)-I(s)|>εd}),εd>0 (3);
if the characteristic point satisfies the formula (3) and M33, judging S as a characteristic point, otherwise, directly excluding the point;
step 23: calculating Hamming distances d between any one feature point S selected from one image and all feature points V in other images, sequencing the obtained distances d, selecting a point with the closest distance as a matching point, and establishing a rough matching point pair to form a feature point set;
step 24: and screening the obtained feature point set by adopting a random sampling consistency method, and finally removing mismatching points to obtain a purified matching feature point set.
4. The method for splicing images with high precision under the multi-machine cooperative constraint as claimed in claim 3, wherein the specific operation steps of step 24 include:
step 241: selecting Q points on the obtained characteristic point set, and calculating all characteristic points P of the first image according to the set registration straight line model1Set of mapped points P on the second image* 2And the mapping relation is satisfied: p* 2 m=f(P1 m) And m is 1,2, …, Q;
step 242: calculating P* 2 mEach point and the second image feature point set P2Euclidean distance of the corresponding point, setting the threshold value of the gray level difference of the characteristic points as delta, and counting P* 2The number N of the characteristic points satisfying that the Euclidean distance is less than deltaT:NT=size(P* 2 m∈{||P2 m-P* 2 m||2<δ}),δ>0;
Step 243: re-randomly selecting Q points, repeating the K steps 241-242, and recording N timesTUntil the iteration is finished;
step 244: selecting N as max (N)Tk(K-1, 2, …, K)) registration model as the final fitting result, thus rejecting the mis-match points.
5. The method for splicing images with high precision under the multi-machine cooperative constraint as claimed in claim 1, wherein the specific operation steps of step 3 include:
step 31: adopting a mobile DLT method to realize linear fitting to the self-adaptive homography matrix according to the coordinate values of the plurality of groups of purified matching target point pairs;
step 32: time sequence correction is carried out on the multi-unmanned aerial vehicle aerial image sequence by adopting the image central point position after homography transformation, and time sequence moving parameter flag of the corresponding imageoThe values of (A) are as follows:
wherein, PoxIs the x coordinate of the central point of the o-th homography transformed image, and L is the width of the reference image;
step 33: traversing each frame of the image sequence according to flagoThe different values of (2) are corrected: when flagoWhen the number of the frames is 1, splicing the frames of the images to be spliced, and when the number of the frames is flagoWhen the image number is-1, the previous frame of the image to be spliced is used for splicing until a flagoWhen the value is 0, outputting the corrected image sequence;
step 34: and recalculating new pixel values of the overlapped area by using the linear gradient of the image, and performing color correction.
6. The method for splicing images with high precision under the multi-machine cooperative constraint as claimed in claim 5, wherein the specific operation steps of step 31 include:
step 311: the feature matching point sets of the two images are respectively X*' and X*Constructing an adaptive homography transformation relation X*'=H*X*;
Step 312: changing the constructed adaptive homography conversion relation into an implicit form, and linearizing into: x*'×H*X*When the value is equal to 0, the following can be obtained:
step 313: memory linearization matrixCan obtain A h 0, for A*Singular value decomposition is carried out to obtain an adaptive homography matrix H*Right singular vector of
Wherein, the weight valueIndicates a control point x*The more recent pixel data is important, the variance σ2Is a measure of how far the measurement point deviates from the control point;
7. The method for stitching high-precision images under multi-machine cooperative constraint as recited in claim 5, wherein the formula for calculating the new pixel value of the overlapped region in step 34 is:
I(x,y)=Il(xl,yl)(1-α)+Ir(xr,yr)α (6),
wherein the content of the first and second substances,represents the distance from the abscissa of the pixel point to the abscissa of the left boundary of the overlap region, (x)l,yl) To stitch the front left image coordinates, (x)r,yr) For stitching front-right image coordinates, Il(xl,yl) For stitching front left image pixels, Ir(xr,yr) To stitch the front right image pixels, (x)lmax,ylmax) The maximum value of the left image coordinates before splicing, (x, y) are the image coordinates after splicing, and I (x, y) is the image pixel after splicing.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110450860.5A CN113313659B (en) | 2021-04-25 | 2021-04-25 | High-precision image stitching method under multi-machine cooperative constraint |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110450860.5A CN113313659B (en) | 2021-04-25 | 2021-04-25 | High-precision image stitching method under multi-machine cooperative constraint |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113313659A true CN113313659A (en) | 2021-08-27 |
CN113313659B CN113313659B (en) | 2024-01-26 |
Family
ID=77371027
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110450860.5A Active CN113313659B (en) | 2021-04-25 | 2021-04-25 | High-precision image stitching method under multi-machine cooperative constraint |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113313659B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114154352A (en) * | 2022-01-18 | 2022-03-08 | 中国科学院长春光学精密机械与物理研究所 | Method and device for designing topological structure of aerial imaging multi-actuator cooperative control |
CN114820737A (en) * | 2022-05-18 | 2022-07-29 | 浙江圣海亚诺信息技术有限责任公司 | Remote sensing image registration method based on structural features |
CN116681590A (en) * | 2023-06-07 | 2023-09-01 | 中交广州航道局有限公司 | Quick splicing method for aerial images of unmanned aerial vehicle |
CN117036666A (en) * | 2023-06-14 | 2023-11-10 | 北京自动化控制设备研究所 | Unmanned aerial vehicle low-altitude positioning method based on inter-frame image stitching |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130208997A1 (en) * | 2010-11-02 | 2013-08-15 | Zte Corporation | Method and Apparatus for Combining Panoramic Image |
WO2014023231A1 (en) * | 2012-08-07 | 2014-02-13 | 泰邦泰平科技(北京)有限公司 | Wide-view-field ultrahigh-resolution optical imaging system and method |
US20190385288A1 (en) * | 2015-09-17 | 2019-12-19 | Michael Edwin Stewart | Methods and Apparatus for Enhancing Optical Images and Parametric Databases |
CN112435163A (en) * | 2020-11-18 | 2021-03-02 | 大连理工大学 | Unmanned aerial vehicle aerial image splicing method based on linear feature protection and grid optimization |
-
2021
- 2021-04-25 CN CN202110450860.5A patent/CN113313659B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130208997A1 (en) * | 2010-11-02 | 2013-08-15 | Zte Corporation | Method and Apparatus for Combining Panoramic Image |
WO2014023231A1 (en) * | 2012-08-07 | 2014-02-13 | 泰邦泰平科技(北京)有限公司 | Wide-view-field ultrahigh-resolution optical imaging system and method |
US20190385288A1 (en) * | 2015-09-17 | 2019-12-19 | Michael Edwin Stewart | Methods and Apparatus for Enhancing Optical Images and Parametric Databases |
CN112435163A (en) * | 2020-11-18 | 2021-03-02 | 大连理工大学 | Unmanned aerial vehicle aerial image splicing method based on linear feature protection and grid optimization |
Non-Patent Citations (1)
Title |
---|
张永梅;张晨希;郭莎;: "基于SIFT特征的彩色图像拼接方法研究", 计算机测量与控制, no. 08 * |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114154352A (en) * | 2022-01-18 | 2022-03-08 | 中国科学院长春光学精密机械与物理研究所 | Method and device for designing topological structure of aerial imaging multi-actuator cooperative control |
CN114154352B (en) * | 2022-01-18 | 2024-05-17 | 中国科学院长春光学精密机械与物理研究所 | Topology structure design method and device for cooperative control of aviation imaging multiple actuators |
CN114820737A (en) * | 2022-05-18 | 2022-07-29 | 浙江圣海亚诺信息技术有限责任公司 | Remote sensing image registration method based on structural features |
CN114820737B (en) * | 2022-05-18 | 2024-05-07 | 浙江圣海亚诺信息技术有限责任公司 | Remote sensing image registration method based on structural features |
CN116681590A (en) * | 2023-06-07 | 2023-09-01 | 中交广州航道局有限公司 | Quick splicing method for aerial images of unmanned aerial vehicle |
CN116681590B (en) * | 2023-06-07 | 2024-03-12 | 中交广州航道局有限公司 | Quick splicing method for aerial images of unmanned aerial vehicle |
CN117036666A (en) * | 2023-06-14 | 2023-11-10 | 北京自动化控制设备研究所 | Unmanned aerial vehicle low-altitude positioning method based on inter-frame image stitching |
CN117036666B (en) * | 2023-06-14 | 2024-05-07 | 北京自动化控制设备研究所 | Unmanned aerial vehicle low-altitude positioning method based on inter-frame image stitching |
Also Published As
Publication number | Publication date |
---|---|
CN113313659B (en) | 2024-01-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108648240B (en) | Non-overlapping view field camera attitude calibration method based on point cloud feature map registration | |
CN110569704B (en) | Multi-strategy self-adaptive lane line detection method based on stereoscopic vision | |
CN113313659B (en) | High-precision image stitching method under multi-machine cooperative constraint | |
CN107063228B (en) | Target attitude calculation method based on binocular vision | |
CN104484648B (en) | Robot variable visual angle obstacle detection method based on outline identification | |
CN114936971A (en) | Unmanned aerial vehicle remote sensing multispectral image splicing method and system for water area | |
CN113532311A (en) | Point cloud splicing method, device, equipment and storage equipment | |
WO2018145291A1 (en) | System and method for real-time location tracking of drone | |
CN111507901B (en) | Aerial image splicing and positioning method based on aerial GPS and scale invariant constraint | |
CN112734841B (en) | Method for realizing positioning by using wheel type odometer-IMU and monocular camera | |
CN110288656A (en) | A kind of object localization method based on monocular cam | |
CN103822616A (en) | Remote-sensing image matching method with combination of characteristic segmentation with topographic inequality constraint | |
CN113850126A (en) | Target detection and three-dimensional positioning method and system based on unmanned aerial vehicle | |
CN113222820B (en) | Pose information-assisted aerial remote sensing image stitching method | |
CN109272574A (en) | Linear array rotary scanning camera imaging model building method and scaling method based on projective transformation | |
CN107677274A (en) | Unmanned plane independent landing navigation information real-time resolving method based on binocular vision | |
CN110021039A (en) | The multi-angle of view material object surface point cloud data initial registration method of sequence image constraint | |
CN109871739B (en) | Automatic target detection and space positioning method for mobile station based on YOLO-SIOCTL | |
CN111060006A (en) | Viewpoint planning method based on three-dimensional model | |
CN113506342B (en) | SLAM omni-directional loop correction method based on multi-camera panoramic vision | |
CN111967337A (en) | Pipeline line change detection method based on deep learning and unmanned aerial vehicle images | |
CN114022560A (en) | Calibration method and related device and equipment | |
Eichhardt et al. | Affine correspondences between central cameras for rapid relative pose estimation | |
CN113624231A (en) | Inertial vision integrated navigation positioning method based on heterogeneous image matching and aircraft | |
CN116740288B (en) | Three-dimensional reconstruction method integrating laser radar and oblique photography |
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 |