CN109785428A - A kind of handheld three-dimensional method for reconstructing based on polymorphic constrained Kalman filter - Google Patents
A kind of handheld three-dimensional method for reconstructing based on polymorphic constrained Kalman filter Download PDFInfo
- Publication number
- CN109785428A CN109785428A CN201910053569.7A CN201910053569A CN109785428A CN 109785428 A CN109785428 A CN 109785428A CN 201910053569 A CN201910053569 A CN 201910053569A CN 109785428 A CN109785428 A CN 109785428A
- Authority
- CN
- China
- Prior art keywords
- polymorphic
- state
- kalman filter
- constrained kalman
- constrained
- 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
Classifications
-
- 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
- Image Analysis (AREA)
- Navigation (AREA)
Abstract
The invention discloses a kind of handheld three-dimensional method for reconstructing based on polymorphic constrained Kalman filter, the close coupling vision inertia odometer based on polymorphic constrained Kalman filter algorithm is introduced, and observes multiple image and delay linearisation in polymorphic constrained Kalman filter algorithm；The polymorphic constrained Kalman filter algorithm includes that state publication and state update, and state publication includes calculating concurrent cloth from IMU vector to do well and covariance matrix, and it includes state enhancing, image procossing, update and condition managing that the state, which updates,.The present invention can be realized the reconstruction of large scene high accuracy three-dimensional.
Description
Technical field
The invention belongs to three-dimensional acquisitions, reconstruction technique field, and in particular to a kind of based on polymorphic constrained Kalman filter
Handheld three-dimensional method for reconstructing can be applied to project Tango equipment.
Background technique
With the arrival in artificial intelligence epoch, in occasions such as robot navigation, unmanned and mappings for three-dimensional reconstruction
Using increasing.The three-dimensional reconstruction system of view-based access control model inertia odometer (Visual-Inertial Odometry, VIO)
It is the critical issue studied at present, and how is applied to three-dimensional reconstruction system on the limited cell phone apparatus of calculation resources also urgently
It solves.VIO will merge the information of vision and Inertial Measurement Unit (Inertial Measurement Unit, IMU), utilize filter
The data fusion that the method for wave carries out multisensor is a kind of common method.The three-dimensional reconstruction system of traditional pure vision exists
Following problems: (1) camera shake or rapid situation of movement under, cause be overlapped between frame and frame less to be difficult to characteristic matching, vision
Information fails in the process, accurately can not carry out three-dimensional reconstruction to model；(2) position is obtained based on several feature single-frame images
Appearance is highly prone to dbjective state variation interference, generates error to attitude algorithm, establishes dislocation Three-dimension Reconstruction Model；(3) tradition three
Dimension reconstructing system algorithm real time linear method causes algorithm complicated, and real-time is poor, more demanding to three-dimensional reconstruction equipment performance.
In recent years, researcher carries out many pilot studys to problem above, and according to the classification method of Kaiser, VIO divides
At based on filtering (filter-based) and based on optimization (optimization-based) two major class.Vision based on filtering
Inertia odometer utilizes the side of maximum a-posteriori estimation (Maximum a posteriori estimation, abbreviation MAP)
Method, the prior distribution of posture and likelihood distribution are established by internal sensor and external sensor measurement respectively.View based on optimization
Feel inertia odometer, using the method for maximal possibility estimation (Maximum likelihood estimation, abbreviation MLE), repeatedly
Batch process, is converted to solving linear algebric equation group by the highest state of total probability for finding to generation measurement.According to feature
It whether include visual signature information in vector, VIO can be divided into loose coupling (loosely-coupled) and close coupling again
(tightly-coupled).Wherein, the loose coupling representative based on optimization is the inertia householder method that Falquez is proposed, this method
The pose transformation that visual odometry obtains is fused in the Optimization Framework of inertia measurement, but this method carries out information fusion
When, best estimate can not be carried out to deviation by handling camera and IMU measurement result respectively, cause error larger.Based on optimization
Close coupling representative is OKVIS, and inertia measurement is tightly integrated into the vision system based on key frame by this method, obtains gravity side
To global coherency and using IMU kinematic motion model robust outlier exclude, compare pure vision and loosely coupled method,
OKVIS shows higher precision and robustness.But OKVIS need establish combine with a unified loss function it is excellent
Change, cause algorithm operation complicated, it is difficult to carry out on the mobile apparatus using.Loose coupling scheme based on filtering is relatively simple,
State vector is not added in the characteristic information of image, but is merged again with inertial data after first handling visual odometry,
Wherein more outstanding is MSF (Multi-Sensor Fusion, Multi-sensor Fusion) that Stephen Weiss et al. is proposed
Method.But the defect of this method is to need to carry out failure detection, causes algorithm to occupy more calculation resources, is not easy to apply
On the mobile apparatus.
Summary of the invention
Goal of the invention of the invention is to provide a kind of handheld three-dimensional method for reconstructing based on polymorphic constrained Kalman filter,
It can be realized the reconstruction of large scene high accuracy three-dimensional.
To achieve the above object of the invention, the technical solution adopted by the present invention is that: one kind be based on polymorphic constrained Kalman filter
Handheld three-dimensional method for reconstructing, introduce the close coupling vision inertia odometer based on polymorphic constrained Kalman filter algorithm, and
Multiple image and delay linearisation are observed in polymorphic constrained Kalman filter algorithm；
The polymorphic constrained Kalman filter algorithm includes that state publication and state update,
State publication includes calculating that concurrent cloth does well and covariance matrix, the state are updated from IMU vector
Including state enhancing, image procossing, update and condition managing.
In above-mentioned technical proposal, state enhancing increases state vector and corresponding with current IMU position and direction
Covariance matrix；
Described image processing extracts corner feature and executes characteristic matching；
The update, each characteristic point completed for track calculate residual vector, and execute mahalanobis distance test,
It is updated using all by the characteristic point of test；
The condition managing deletes all characteristic points from state vector and calculates the IMU state completed.
In above-mentioned technical proposal, observed in the polymorphic constrained Kalman filter algorithm using sliding window method more
Frame image and delay linearisation.
Due to the above technical solutions, the present invention has the following advantages over the prior art:
1. the present invention is overcome camera and is trembled by the close coupling vision inertia odometer of the polymorphic constrained Kalman filter of introducing
The problem of dynamic, scene modeling characteristic point is few and state change etc. is brought, and in polymorphic constrained Kalman filter algorithm, it is innovative
Observation multiple image and delay linearisation, improve three-dimensional reconstruction precision and efficiency；
2. the present invention is tested on KITTI and Starry Night data set, three maintenance and operations of the invention are effectively verified
The precision of motion tracking, and the three-dimensional reconstruction of progress different scenes is tested on project Tango mobile phone, by being with other
System is compared and analyzes, and demonstrating the present invention may be implemented the reconstruction of large scene high accuracy three-dimensional, is better than current existed system.
Detailed description of the invention
Fig. 1 is the structural schematic diagram of the polymorphic constrained Kalman filter algorithm of the embodiment of the present invention one.
Fig. 2 is the Starry Night data set test result schematic diagram of the embodiment of the present invention one.
Fig. 3 is the Starry Night data set test analysis schematic diagram of the embodiment of the present invention one.
Fig. 4~12 are the three-dimensional reconstruction effect contrast figures of the embodiment of the present invention one.
Specific embodiment
The invention will be further described with reference to the accompanying drawings and embodiments:
Embodiment one:
Shown in Figure 1, a kind of handheld three-dimensional method for reconstructing based on polymorphic constrained Kalman filter is introduced based on more
The close coupling vision inertia odometer of modal constraint Kalman filtering algorithm, and observation is more in polymorphic constrained Kalman filter algorithm
Frame image and delay linearisation；
The polymorphic constrained Kalman filter algorithm includes that state publication and state update,
State publication includes calculating that concurrent cloth does well and covariance matrix, the state are updated from IMU vector
Including state enhancing, image procossing, update and condition managing.
In the present embodiment, the state enhancing increases state vector and corresponding association side with current IMU position and direction
Poor matrix；
Described image processing extracts corner feature and executes characteristic matching；
The update, each characteristic point completed for track calculate residual vector, and execute mahalanobis distance test,
It is updated using all by the characteristic point of test；
The condition managing deletes all characteristic points from state vector and calculates the IMU state completed.
In the present embodiment, observation multiframe figure is carried out using sliding window method in the polymorphic constrained Kalman filter algorithm
Picture and delay linearisation.
Specifically, the present invention constructs the VIO system of a joint VO and IMU, and a coordinate system { I } is attached to IMU,
And track movement of this frame relative to global frame.
Standard convention is followed, the IMU state vector at time l is defined as 16 × 1 vectors, whereinIt is unit quaternary
Number indicates at time l from global frame to the rotation of IMU frame,GplWithGvlIt is the IMU position and speed in global frame respectively, and
bglAnd balIt is gyroscope and accelerometer deviation respectively.
As shown in formula (1):
IMU bias state is defined using the standard increment error of position, speed and deviation.Using quaternary numberTo describe
The true difference between estimation direction, it is shown in such as formula (2) to define azimuthal error:
WhereinIt is quaternary number multiplication.Intuitively,It is that the global frame of estimation is made to match required rotation with true frames
Turn.It is indicated to obtain the minimum of this rotation, it was noted thatIt can be write as:
WhereinIt is 3 × 1 vectors for describing the deflection error about three axis.It is defined by above-mentioned error, IMU mistake
State is defined as 15 × 1 vectors:
MSCKF of the invention is a kind of innovatory algorithm based on EKF, is maintained in state vector based on sliding window
Posture, and apply probability constraints between these postures using observation of characteristics.The state vector of MSCKF at time l is defined
Are as follows:
Wherein,It is for time i=l-N, L, at l-1, when last N number of image of record
IMU posture.
In MSCKF algorithmic procedure, IMU state measurement is used for IMU state estimation and filtering covariance matrix calculates.When
When camera record new images, MSCKF state and covariance just will increase the copy of current IMU posture, and to image at
Reason is to extract and matching characteristic.Each feature is tracked until its all measured value is made available by, all measurements of use at this time are simultaneously
It is updated.
In renewal equation, the spy arrived using the N number of image viewing recorded in MSCKF state vector is updated at time l
Levy fiThe case where.The first step of the process is to obtain feature locationsEstimated value, therefore it is logical using the measured value of all features
Gauss-Newton is crossed to minimize to estimate its position.Finally, residual error when calculating j=l-N ..., l-1 in formula (6):
WhereinWithIt is the current evaluated error and feature locations error of j-th of posture respectively；MatrixWithIt is corresponding Jacobian matrix, byWithAssessment obtains.BecauseBy j=l-N ..., l-1'sWith
zijIt calculates, the feature locations error of the residual error in formula (7)WithAnd nijIt is all related.Therefore, in MSCKF,
Continuation is removed from residual equationFirstly, establishing the vector comprising N number of residual error from all pattern measurements:
Wherein, riAnd niIt is element r respectivelyijAnd nijBlock vector,WithIt is block rowWithMatrix.With
Afterwards, residual vector is definedWherein ViIt is a matrix, column are constitutedLeft side kernel basis.By public affairs
Formula (9) can obtain:
Wherein,And nothingThe residual vector r in formula (10)i oWith the error in characteristic coordinates without
It closes.In order to improve efficiency of algorithm, ri oWithV is clearly formed noiIn the case where i.e. calculated.Once calculating ri oWithJust to residual error ri oCarry out Mahalanobis gate test.Specifically, it calculates
And by its with by the χ with 2N-3 freedom degree2The threshold value that provides of the 95th percentile of distribution is compared that (2N-3 is
Residual vector ri oIn element quantity).
If the function continues to use r by testi oAnd it is carried out by the residual error of the every other function of gate test
It updates.After this more kainogenesis, i.e., those postures for having been used for completing to update observed are deleted from sliding window.
The present invention constrains using sliding window and postpones the method for linearisation, can make the Three-dimensional Gravity based on MSCKF
It is more accurate to build system data fusion；Condition managing leaves out redundancy in time, and a large amount of saving fortune are deposited and memory, makes to be based on
The three-dimensional reconstruction system of MSCKF is more applicable for the limited mobile device of computing resource.
In order to verify the validity of inventive algorithm, the present invention is respectively provided in KITTI and Starry Night data
Emulation testing is carried out to inventive algorithm on collection and carries out three-dimensional reconstruction experimental evaluation on project Tango platform.?
In KITTI data set and the experiment of Starry Night data set, Experimental Hardware is configured to the Intel Core i7- of memory 8G
The desktop computer of 4790 CPU, the software environment used are that the MATLAB R2017a under ubuntu 14.04 is tested.Three
Dimension is rebuild in experiment, and source code, which develops environment, to be carried out in the Eclipse 3.8.1 of ubuntu 16.04, is packaged into later
The APK file installation and operation of 48.6M is in project Tango equipment.
The experiment porch based on project Tango equipment is built first, and the present invention is flat using project Tango is carried
Association's PHB2 mobile phone of platform is as experiment porch.Associate PHB2 mobile phone and uses valiant imperial 652 processor of 8 core 1.8GHz high passes, 4G fortune
It deposits, 64G memory, Android version is 6.0.1, and Tango version is 1.2.The mobile phone has comprising wide visual field fisheye camera, depth camera
And color camera.Fisheye camera visual angle is close to 180 degree, using global shutter working method.Colour imagery shot is in present system
In be not used to motion tracking, and be used for give Three-dimension Reconstruction Model rendering color.Depth camera is based on infrared structure light, acquisition
To depth image resolution ratio be QVGA (Quarter Video Graphics Array), dense degree and quality are similar to
Kinect.In addition to this, which also has the Inertial Measurement Unit of low-power consumption, by 3 axis accelerometers and 3 axis gyroscope groups
At.The dense arrangement mode that the system is based on hash data structure (hash) can be in limited calculating using improved TSDF algorithm
Real-time three-dimensional is completed under resource to rebuild and do not have to GPU and accelerate.Wherein, the core of system is in the vision inertia based on MSCKF
Journey meter.
In order to effectively verify the validity of the system based on MSCKF algorithm, selected Starry Night data set and
For KITTI data set to the present invention is based on the systems of MSCKF algorithm to test, two datasets all contain binocular solid camera
It with IMU information, is used since present system is designed on project Tango mobile phone, fisheye camera is used only in VIO
And IMU, so the left camera image and IMU information of Usage data collection.
It is shown in Figure 2, it is the sheet based on MSCKF algorithm of the different characteristic point quantity under Starry Night data set
Invention system and the system contrast schematic diagram that MSCKF fusion visual information is not added.It is 40,60,80 and 100 in characteristic point quantity
Scene under, analyze present system and be not added the system of MSCKF rotation root-mean-square error and displacement root-mean-square error.
It can be seen that the rotation root-mean-square error and displacement root-mean-square error of present system are much smaller than in different characteristic point
The system of MSCKF algorithm is not added.
It is shown in Figure 3, calculate within 1415 to 1715 period the average rotation root-mean-square error of two systems and
Average displacement root-mean-square error.It can be seen that present system is in the case of characteristic point more crypto set, the average rotation of system
Square root error and average displacement square root error are smaller.Compare the system that MSCKF is not added, the average rotation of present system
Square root error and average displacement square root error are smaller, show that system-computed pose is more accurate, compare existing system,
Present system has lesser error under various feature dot densities, has stronger robustness and stability.Carrying out three
When dimension is rebuild, it is possible to prevente effectively from the accuracy of model also can more increase phenomena such as drift dislocation.
In order to effectively verify the stability and advantage of present system on the mobile apparatus, Open Constructor is chosen
It can be applied to the system in Project Tango equipment as a comparison with RTAB-Map two.Fig. 4, Fig. 7 and Figure 10 are the present invention
The three-dimensional reconstruction effect picture of system, Fig. 5, Fig. 8 and Figure 11 are the three-dimensional reconstruction effect picture of RTAB-Map system, Fig. 6, Fig. 9 and figure
12 be the three-dimensional reconstruction effect picture of Open Constructor system.It can be seen from the figure that RTAB-Map system is facing glass
When the visions such as glass, white wall are difficult to matched frame, easily there is the case where closed loop failure, so as to cause pose drift, model dislocation；
In Open Constructor system, in camera shake, pose variation can not be preferably handled, the model established out is caused
Easily torsional deformation, and serious distortion is described in part.Compared to both the above system, it can be seen that in a system of the invention, be based on
The VIO of MSCKF solves precisely pose, can preferably cope with the sparse situation of characteristic point.In localised jitter, phase is rotated rapidly
Machine and it is upper downstairs in the case of, present system to pose solve precisely, still can establish the preferable threedimensional model of effect.
It is corresponding from above each Three-dimension Reconstruction Model to choose 20 Eigenvectors progress error analyses in table 1, by feature
The true value of line segment is compared with measured value, calculates error rate.Wherein, Fig. 5 experimental error is excessive, does not establish out corresponding
Threedimensional model.Present system error is much smaller than RTAB-Map and Open Constructor, in terms of modeling and optimizing the time,
In the case where the overall modeling optimization time is shorter, lower error rate is reached, has improved the accuracy and efficiency of threedimensional model.
The comparison of 1 experimental result of table
By being tested above it can be proved that the three dimension system tracking trajectory capacity proposed by the present invention based on MSCKF is stronger,
Three-dimensional pose calculates precisely, and stability is strong.Under the various scenes on project Tango mobile phone three-dimensional reconstruction experiment and
Compare homogeneous system, it can be seen that present system three-dimensional reconstruction effect is better than homogeneous system, and is suitble to large scene high-precision three
Dimension is rebuild.
Claims (3)
1. a kind of handheld three-dimensional method for reconstructing based on polymorphic constrained Kalman filter, it is characterised in that: introduce based on polymorphic
The close coupling vision inertia odometer of constrained Kalman filter algorithm, and multiframe is observed in polymorphic constrained Kalman filter algorithm
Image and delay linearisation；
The polymorphic constrained Kalman filter algorithm includes that state publication and state update,
State publication includes calculating that concurrent cloth does well and covariance matrix, the state update include from IMU vector
State enhancing, image procossing, update and condition managing.
2. the handheld three-dimensional method for reconstructing according to claim 1 based on polymorphic constrained Kalman filter, feature exist
In: the state enhancing increases state vector and corresponding covariance matrix with current IMU position and direction；
Described image processing extracts corner feature and executes characteristic matching；
The update, each characteristic point completed for track calculate residual vector, and execute mahalanobis distance test, use
It is all to be updated by the characteristic point of test；
The condition managing deletes all characteristic points from state vector and calculates the IMU state completed.
3. the handheld three-dimensional method for reconstructing according to claim 1 based on polymorphic constrained Kalman filter, feature exist
In: observation multiple image is carried out using sliding window method in the polymorphic constrained Kalman filter algorithm and delay linearizes.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910053569.7A CN109785428B (en) | 2019-01-21 | 2019-01-21 | Handheld three-dimensional reconstruction method based on polymorphic constraint Kalman filtering |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910053569.7A CN109785428B (en) | 2019-01-21 | 2019-01-21 | Handheld three-dimensional reconstruction method based on polymorphic constraint Kalman filtering |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109785428A true CN109785428A (en) | 2019-05-21 |
| CN109785428B CN109785428B (en) | 2023-06-23 |
Family
ID=66500961
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910053569.7A Active CN109785428B (en) | 2019-01-21 | 2019-01-21 | Handheld three-dimensional reconstruction method based on polymorphic constraint Kalman filtering |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109785428B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110262521A (en) * | 2019-07-24 | 2019-09-20 | 北京智行者科技有限公司 | A kind of automatic Pilot control method |
| CN110340887A (en) * | 2019-06-12 | 2019-10-18 | 西安交通大学 | A method of the oiling robot vision guide based on image |
| CN113674412A (en) * | 2021-08-12 | 2021-11-19 | 浙江工商大学 | Pose fusion optimization-based indoor map construction method and system and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107656545A (en) * | 2017-09-12 | 2018-02-02 | 武汉大学 | A kind of automatic obstacle avoiding searched and rescued towards unmanned plane field and air navigation aid |
| CN108629793A (en) * | 2018-03-22 | 2018-10-09 | 中国科学院自动化研究所 | The vision inertia odometry and equipment demarcated using line duration |
| CN109029448A (en) * | 2018-06-28 | 2018-12-18 | 东南大学 | The IMU of monocular vision inertial positioning assists trace model |
-
2019
- 2019-01-21 CN CN201910053569.7A patent/CN109785428B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107656545A (en) * | 2017-09-12 | 2018-02-02 | 武汉大学 | A kind of automatic obstacle avoiding searched and rescued towards unmanned plane field and air navigation aid |
| CN108629793A (en) * | 2018-03-22 | 2018-10-09 | 中国科学院自动化研究所 | The vision inertia odometry and equipment demarcated using line duration |
| CN109029448A (en) * | 2018-06-28 | 2018-12-18 | 东南大学 | The IMU of monocular vision inertial positioning assists trace model |
Non-Patent Citations (3)
| Title |
|---|
| ANASTASIOS I. MOURIKIS等: "A Multi-State Constraint Kalman Filter for Vision-aided Inertial Navigation", 《2007 IEEE INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION》 * |
| TONG QIN等: "VINS-Mono: A Robust and Versatile Monocular Visual-Inertial State Estimator", 《IEEE TRANSACTIONS ON ROBOTICS》 * |
| 万众: "微小型飞行器立体视觉避障与自主导航关键技术研究", 《万方》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110340887A (en) * | 2019-06-12 | 2019-10-18 | 西安交通大学 | A method of the oiling robot vision guide based on image |
| CN110262521A (en) * | 2019-07-24 | 2019-09-20 | 北京智行者科技有限公司 | A kind of automatic Pilot control method |
| CN113674412A (en) * | 2021-08-12 | 2021-11-19 | 浙江工商大学 | Pose fusion optimization-based indoor map construction method and system and storage medium |
| CN113674412B (en) * | 2021-08-12 | 2023-08-29 | 浙江工商大学 | Pose fusion optimization-based indoor map construction method, system and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109785428B (en) | 2023-06-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109307508B (en) | Panoramic inertial navigation SLAM method based on multiple key frames | |
| Qin et al. | Vins-mono: A robust and versatile monocular visual-inertial state estimator | |
| CN109993113B (en) | Pose estimation method based on RGB-D and IMU information fusion | |
| CN111156984B (en) | Monocular vision inertia SLAM method oriented to dynamic scene | |
| CN108062776B (en) | Camera Attitude Tracking method and apparatus | |
| Newcombe et al. | Kinectfusion: Real-time dense surface mapping and tracking | |
| CN104658012B (en) | Motion capture method based on inertia and optical measurement fusion | |
| CN109166149A (en) | A kind of positioning and three-dimensional wire-frame method for reconstructing and system of fusion binocular camera and IMU | |
| US20220051031A1 (en) | Moving object tracking method and apparatus | |
| CN113052908B (en) | Mobile robot pose estimation algorithm based on multi-sensor data fusion | |
| CN108682027A (en) | VSLAM realization method and systems based on point, line Fusion Features | |
| CN109540126A (en) | A kind of inertia visual combination air navigation aid based on optical flow method | |
| CN110726406A (en) | Improved nonlinear optimization monocular inertial navigation SLAM method | |
| CN111862316B (en) | Three-dimensional reconstruction method of dense direct RGBD (Red Green blue-white) of tight coupling of IMU (inertial measurement Unit) based on optimization | |
| JP6483832B2 (en) | Method and system for scanning an object using an RGB-D sensor | |
| Li et al. | Large-scale, real-time 3D scene reconstruction using visual and IMU sensors | |
| CN112556719B (en) | Visual inertial odometer implementation method based on CNN-EKF | |
| CN112815939A (en) | Pose estimation method for mobile robot and computer-readable storage medium | |
| CN109785428A (en) | A kind of handheld three-dimensional method for reconstructing based on polymorphic constrained Kalman filter | |
| CN115371665B (en) | Mobile robot positioning method based on depth camera and inertial fusion | |
| CN104848861A (en) | Image vanishing point recognition technology based mobile equipment attitude measurement method | |
| CN114485640A (en) | Monocular vision inertia synchronous positioning and mapping method and system based on point-line characteristics | |
| CN111595332B (en) | Full-environment positioning method integrating inertial technology and visual modeling | |
| CN116619358A (en) | Self-adaptive positioning optimization and mapping method for autonomous mining robot | |
| Huttunen et al. | A monocular camera gyroscope |
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 |