CN104732587B - A kind of indoor 3D semanteme map constructing method based on depth transducer - Google Patents
A kind of indoor 3D semanteme map constructing method based on depth transducer Download PDFInfo
- Publication number
- CN104732587B CN104732587B CN201510175129.0A CN201510175129A CN104732587B CN 104732587 B CN104732587 B CN 104732587B CN 201510175129 A CN201510175129 A CN 201510175129A CN 104732587 B CN104732587 B CN 104732587B
- Authority
- CN
- China
- Prior art keywords
- image
- rgb
- map
- indoor
- acquisition
- 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.)
- Active
Links
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a kind of indoor 3D semanteme map constructing method based on depth transducer, this method comprises: using the color depth RGB-D image of depth transducer acquisition indoor environment, and indoor 3D map is constructed with this；Collected RGB-D image is split, and calculates the color and shape feature of RGB-D image after segmentation, obtains corresponding semantic information；The semantic information of acquisition is merged with interior 3D map, obtains interior 3D semanteme map.By using method disclosed by the invention, can construct comprising the semantic informations such as structure semantics information and furniture semantic information, so that robot executes high-rise intelligent operation.
Description
Technical field
The present invention relates to robot vision scene understanding technical field more particularly to a kind of interiors based on depth transducer
3D semanteme map constructing method.
Background technique
The Semantic Aware of robot is the core and vital technology of indoor service robot.Traditional robot
Indoor map is established with limitation by laser.The map that one side laser is established is 2D map, since it lacks effectively
3D information can only hide ground object, the object with certain altitude can not be hidden when carrying out mobile avoidance.It is another
Aspect, the map that robot can only be established using laser carry out some bottom operations, and such as avoidance is mobile, path planning.And machine
People does not get a real idea of ambient enviroment.For household service robot, environment is got a real idea of, understands that the demand of user is heavy to closing
Target want and artificial intelligence is that intelligent robot is different from one of key factor of industrial robot.
In recent years with the development of depth transducer (for example, kinect etc.), the indoor 3D sensing capability of robot is gradually
Enhancing, in academia, a series of 3D indoor environment map constructing methods come into being, such as RGB-D Mapping, Kinect
Fusion etc.." ATAP " department under industry, Google discloses newest development project Project Tango, should
Project is researching and developing a kind of smart phone of band 3D environment induction technology, is expected to that real-time 3D environment can be established at any time in future
Structure chart.However, these technologies all establish interior 3D environmental structure map just with depth sensing capability, lack in map
Where understanding semantically is such as metope, where is desk etc..
Summary of the invention
The object of the present invention is to provide a kind of indoor 3D semanteme map constructing method based on depth transducer, can construct
Comprising the semantic informations such as structure semantics information and furniture semantic information, so that robot executes high-rise intelligent operation.
The purpose of the present invention is what is be achieved through the following technical solutions:
A kind of indoor 3D semanteme map constructing method based on depth transducer, this method comprises:
Using the color depth RGB-D image of depth transducer acquisition indoor environment, and indoor 3D map is constructed with this；
Collected RGB-D image is split, and calculates the color and shape feature of RGB-D image after segmentation, is obtained
Obtain corresponding semantic information；
The semantic information of acquisition is merged with interior 3D map, obtains interior 3D semanteme map.
Further, the color depth RGB-D image using depth transducer acquisition indoor environment, and constructed with this
Indoor 3D map includes:
By the hand-held equipment equipped with depth transducer of user or the mobile robot by being equipped with depth transducer to indoor ring
Border is scanned, and obtains continuous RGB-D image；
Rotational translation matrix between the continuous RGB-D image of pretreatment acquisition is carried out to the continuous RGB-D image, from
And continuous RGB-D image mosaic is constructed to indoor 3D map.
Further, the rotation continuous RGB-D image carried out between the continuous RGB-D image of pretreatment acquisition
Turn translation matrix to include the following steps:
The angle point in each frame RGB-D image is calculated, and the angle point calculated is tracked by optical flow algorithm；
All consistent angle points in adjacent two frames RGB-D image are found out using the consistent RANSAC algorithm of random sampling, as
Characteristic point pair, remaining angle point filter out；
According to the Distance Judgment interframe distance of characteristic point pair in adjacent two frames RGB-D image, if it is determined that interframe distance is greater than
Preset value, then it represents that angle point lazy weight, then calculate in RGB-D image ORB characteristic point and description, and according to description come
Calculate the ORB characteristic point pair of adjacent two frames RGB-D image；
It is filtered processing using ORB characteristic point of the RANSAC algorithm to acquisition, following formula is recycled to calculate adjacent two
The rotational translation matrix of frame RGB-D image:
Wherein, n is the number of adjacent two frames RGB-D image characteristic point pair, piWith qiRespectively adjacent two frames RGB-D image
Middle ith feature point, R, T are respectively rotation, translation matrix；To make E is R, T of minimum value as final result.
Further, it is described by continuous RGB-D image mosaic come before constructing indoor 3D map further include: to each frame
RGB-D image carries out LS-SVM sparseness.
Further, described that collected RGB-D image is split, and calculate the color of RGB-D image after segmentation
With shape feature, obtaining corresponding semantic information includes:
Super-pixel segmentation is carried out respectively to collected each frame RGB-D image using 2D image segmentation algorithm；
The color and shape feature of each super-pixel are calculated using computer vision algorithms make, recycle support vector machines
Classifier classifies to the color calculated with shape feature, obtains the classifier of several semantic classes；
Classified using the semantic classes classifier of acquisition to the super-pixel in each frame RGB-D image, to obtain
The semantic information of RGB-D image.
Further, the semantic information by acquisition is merged with indoor 3D map, obtains interior 3D semanteme map
Include:
Indoor 3D map is divided into several voxels；
Using the Dense CRF model in machine learning and the semantic information of the RGB-D image obtained is combined to carry out voxel
Semantic information reasoning determines the semantic information of each voxel, forms final semantic 3D scene map.
As seen from the above technical solution provided by the invention, the field 3D of environment can be established by scanning ambient enviroment
Scape map, while automatic understanding ambient enviroment add semantic tagger (wall, ground, ceiling, sofa, chair etc.) to environment, most
Indoor 3D semanteme map is obtained eventually；So that intelligent robot, the 3D environmental map being not only able to around establishing carries out bottom
Layer operation (mobile, avoidance etc.), while being also understood that ambient enviroment, reach the real purpose of intelligent semantic perception.
Detailed description of the invention
In order to illustrate the technical solution of the embodiments of the present invention more clearly, required use in being described below to embodiment
Attached drawing be briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention, for this
For the those of ordinary skill in field, without creative efforts, it can also be obtained according to these attached drawings other
Attached drawing.
Fig. 1 is a kind of stream of the indoor 3D semanteme map constructing method based on depth transducer provided in an embodiment of the present invention
Cheng Tu；
Fig. 2 is the flow chart of RGB-D image preprocessing process provided in an embodiment of the present invention.
Specific embodiment
With reference to the attached drawing in the embodiment of the present invention, technical solution in the embodiment of the present invention carries out clear, complete
Ground description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.Based on this
The embodiment of invention, every other implementation obtained by those of ordinary skill in the art without making creative efforts
Example, belongs to protection scope of the present invention.
Fig. 1 is a kind of stream of the indoor 3D semanteme map constructing method based on depth transducer provided in an embodiment of the present invention
Cheng Tu.As shown in Figure 1, this method mainly includes the following steps:
Step 11, the RGB-D image that indoor environment is acquired using depth transducer, and indoor 3D map is constructed with this.
It, can be by the hand-held equipment (for example, kinect) for being equipped with depth transducer of user or by setting in the embodiment of the present invention
There is the mobile robot of depth transducer to be scanned indoor environment, obtains continuous RGB-D (color depth) image；It is right again
The continuous RGB-D image carries out the rotational translation matrix between the continuous RGB-D image of pretreatment acquisition, thus by continuous
RGB-D image mosaic constructs indoor 3D map.
As described in Figure 2, RGB-D image preprocessing process mainly includes the following steps:
Angle point in step 21, each frame RGB-D image of calculating, and the angle point calculated is tracked by optical flow algorithm.
Step 22 finds out all consistent angle points in adjacent two frames RGB-D image using RANSAC algorithm, as characteristic point
Right, remaining angle point filters out.
It alternates betwwen good and bad by the angle point that step 21 tracks, in the embodiment of the present invention, utilizes RANSAC (random sampling is consistent)
Algorithm finds out all consistent angle points in adjacent two frames RGB-D image, as characteristic point pair.
Illustratively, three characteristic points pair can be randomly choosed, calculate rotational transformation matrix T, while statistics meets matrix T's
Characteristic point is to number.It repeats the above process repeatedly, the characteristic point T most to number is exactly best rotational transformation matrix best
T is unsatisfactory for the characteristic point of best T to being exactly bad tracking, needs to filter out.
Step 23, according to the Distance Judgment interframe distance of characteristic point pair in adjacent two frames RGB-D image, if it is determined that frame pitch
From being greater than preset value, then it represents that angle point lazy weight, then calculate in RGB-D image ORB characteristic point and description, and according to retouching
Son is stated to calculate the ORB characteristic point pair of adjacent two frames RGB-D image.
In the embodiment of the present invention, if interframe apart from too small, illustrates that camera is not substantially dynamic, it is not added into global scene
In；If interframe distance is greater than preset value (can according to the actual situation or experience setting), then it represents that angle point quantity may be not enough；This
When, it needs using other characteristic points, for example, SURF (Speeded Up Robust Features accelerates robust feature), ORB
(ORiented Brief)。
In the embodiment of the present invention, it is contemplated that the reason of efficiency, using ORB characteristic point；OpenCV (vision open source can be passed through
Library) ORB cuda (the GPU algorithm of ORB characteristic point) calculate ORB characteristic point and description son, further according to description son, to calculate
The ORB characteristic point pair of adjacent two frames RGB-D image.
Step 24 is filtered processing using ORB characteristic point of the RANSAC algorithm to acquisition, recycles following formula
(Point to Point ICP, with regard to close-point search method) calculates the rotational translation matrix of adjacent two frames RGB-D image:
Wherein, n is the number (characteristic point pair and mistake obtained including step 22 of adjacent two frames RGB-D image characteristic point pair
Filter post-processing after ORB characteristic point to), piWith qiIth feature point is (that is, p in respectively adjacent two frames RGB-D imageiWith qi
For ith feature point to), R, T be respectively rotate, translation matrix；To make E is R, T of minimum value as final result.
Above-mentioned formula has analytic solutions.Common method has: singular value decomposition, quaternary number, orthogonal matrix method, biquaternion side
Method, the method that singular value decomposition can be used in the present embodiment, and by its parallelization.
Further, consider limited memory, the larger reason of image data occupied space, by continuous RGB-D image mosaic
It also needs to carry out LS-SVM sparseness to each frame RGB-D image before to construct indoor 3D map.
Illustratively, the image point cloud data of a 640*480 will occupy the memory of 3.5M, in order to save space, processing
More images require only one point (that is, resolution ratio is 1cm) in the cube of every 1cm in this example.
In this example, rarefaction is carried out to the space of 3m*3m*3m, since resolution ratio is 1cm, so needing 300*300*
The state array of 300 sizes, at the end GPU, per thread handles a point in image, calculates it relative to state according to coordinate
The index of array judges whether space is occupied according to state array, does not occupy labeled as occupying, occupies, which is not added
Enter in global scene.
On the other hand, the above-mentioned treatment process of the embodiment of the present invention can be realized based on GPU (graphics processor), relative to
It can get faster treatment effeciency for CPU.
Illustratively, when executing abovementioned steps 22, GPU algorithm is as follows: for the end GPU, utilizing each block (module) mould
Intend a RANSAC iteration, 250 iteration are exactly 250 block, and a thread randomly chooses three characteristic points pair in block,
T is calculated, whether per thread statistics meets T in block later, counts number.The last most T of reselection number are as finally
T.The random function API of cuda (universal parallel computing architecture), atomic operation, shared drive, parallel specification etc. have been used herein
Mechanism.
When executing the newly-increased ORB characteristic point processing in abovementioned steps 23, at the end GPU, looked for using the method for parallel specification
Two closest correspondences of each characteristic point pair；The sum operation for becoming parallel specification is needed in this method only to take minimum operation
?；Meanwhile if closest characteristic point distance is less than 0.8 times of secondary neighbouring characteristic point distance, it is believed that this is
Good matching, this process are only needed with per thread to each characteristic point to parallel processing.
When executing calculating rotational translation matrix process in abovementioned steps 24,3* the end GPU: is calculated using the method for parallel specification
H is carried out singular value decomposition, H=U*V later at the end CPU by 3 matrix Hst, R=V*Ut, T=p '-R*p, such spin matrix R,
Translation matrix T, which is just solved, to be come；The correlation formula at the end GPU are as follows:
piWith qiIth feature point is (that is, p in respectively adjacent two frames RGB-D imageiWith qiFor ith feature point to),
R, T is respectively rotation, translation matrix, which is one of method for calculating rotational translation matrix, i.e. singular value decomposition side
Method.
Intuitively to embody based on GPU and the efficiency based on CPU processing, the present embodiment has also carried out experiment and has compared, and compares
The results are shown in Table 1.
| Method | CPU | GPU |
| Corner | 36 | 14.15 |
| Corner pair | 21 | 12.8 |
| RANSAC | 40 | 3.6 |
| ORB | 118 | 24.5 |
| Find ORB pair | 44 | 1.43 |
| Point to Point ICP | 12 | 0.08 |
| Filter | 162 | 15 |
1 comparison result of table
From table 1 it follows that the program at the end GPU is faster relative to the end CPU program.CPU program is per second to handle 4.12
Frame data, GPU program is per second to handle 12.5 frame data；Available 3 times of acceleration is handled based on GPU.
Step 12 is split collected RGB-D image, and calculates the color and shape of RGB-D image after segmentation
Feature obtains corresponding semantic information.
For every frame RGB-D image, its semantic segmentation and mark (semantic are generated in the embodiment of the present invention
segmentation).Specifically:
Firstly, carrying out super-pixel respectively to collected each frame RGB-D image using 2D image segmentation algorithm
(superpixel) divide；
Then, it needs to carry out semantic classification to super-pixel.Semantic classes includes structured sort (wall, ground etc.) and household
Classification (desk, chair, bed etc.).Illustratively, it can use the semanteme point that SVM (support vector machines) classifier realizes super-pixel
Class.The color and shape feature of each super-pixel are calculated first with computer vision algorithms make, and SVM classifier is recycled to arrive calculating
Color classify with shape feature, obtain the classifier of several semantic classes；
Finally, the semantic classes classifier using acquisition classifies to the super-pixel in each frame RGB-D image, thus
Obtain the semantic information of RGB-D image.
Step 13 merges the semantic information of acquisition with interior 3D map, obtains interior 3D semanteme map.
Firstly, indoor 3D map is divided into several voxels (voxel).
Then, semantic information is added to each voxel, in the present embodiment, utilizes the Dense CRF (depth in machine learning
Condition random field) model and combine obtain RGB-D image semantic information carry out the reasoning of voxel semantic information；Wherein, Dense
Node potential energy (unary potential) in CRF can be obtained by the semantic information of RGB-D picture, be associated with potential energy
(pairwise potential) can be defined by Gaussian function；Pass through the reasoning of Dense CRF model, it may be determined that each voxel
Semantic information, form final semantic 3D scene map.
The 3D scene map of environment can be established by scanning ambient enviroment in the scheme of the embodiment of the present invention, while automatic
Understand ambient enviroment, semantic tagger (wall, ground, ceiling, sofa, chair etc.) is added to environment, finally obtains indoor 3D
Semantic map；So that intelligent robot, the 3D environmental map being not only able to around establishing carries out bottom operation, and (movement, is kept away
Barrier etc.), while being also understood that ambient enviroment, reach the real purpose of intelligent semantic perception.
Through the above description of the embodiments, those skilled in the art can be understood that above-described embodiment can
The mode of necessary general hardware platform can also be added to realize by software by software realization.Based on this understanding,
The technical solution of above-described embodiment can be embodied in the form of software products, which can store non-easy at one
In the property lost storage medium (can be CD-ROM, USB flash disk, mobile hard disk etc.), including some instructions are with so that a computer is set
Standby (can be personal computer, server or the network equipment etc.) executes method described in each embodiment of the present invention.
The foregoing is only a preferred embodiment of the present invention, but scope of protection of the present invention is not limited thereto,
Within the technical scope of the present disclosure, any changes or substitutions that can be easily thought of by anyone skilled in the art,
It should be covered by the protection scope of the present invention.Therefore, protection scope of the present invention should be with the protection model of claims
Subject to enclosing.
Claims (3)
1. a kind of indoor 3D semanteme map constructing method based on depth transducer, which is characterized in that this method comprises:
Using the color depth RGB-D image of depth transducer acquisition indoor environment, and indoor 3D map is constructed with this；
Collected RGB-D image is split, and calculates the color and shape feature of RGB-D image after segmentation, acquisition pair
The semantic information answered；
The semantic information of acquisition is merged with interior 3D map, obtains interior 3D semanteme map；
Wherein, the color depth RGB-D image using depth transducer acquisition indoor environment, and with constructing indoor 3D with this
Figure includes: by the hand-held equipment equipped with depth transducer of user or the mobile robot by being equipped with depth transducer to indoor environment
It is scanned, obtains continuous RGB-D image；Pretreatment is carried out to the continuous RGB-D image and obtains continuous RGB-D image
Between rotational translation matrix, so that continuous RGB-D image mosaic to be constructed to indoor 3D map；
Also, before continuous RGB-D image mosaic to construct to indoor 3D map further include: carried out to each frame RGB-D image
LS-SVM sparseness, process are as follows: assuming that only one point in the cube of every 1cm, that is, resolution ratio 1cm；To 3m*3m*
The space of 3m carries out rarefaction, and resolution ratio 1cm then needs the state array of 300*300*300 size, at the end GPU, each
A point in thread process image calculates its index relative to state array according to coordinate, is judged according to state array empty
Between it is whether occupied, do not occupy labeled as occupy, occupy then will the point be added global scene in；
The rotational translation matrix packet continuous RGB-D image carried out between the continuous RGB-D image of pretreatment acquisition
Include following steps:
The angle point in each frame RGB-D image is calculated, and the angle point calculated is tracked by optical flow algorithm；
All consistent angle points in adjacent two frames RGB-D image are found out using the consistent RANSAC algorithm of random sampling, as feature
Point pair, remaining angle point filters out；
According to the Distance Judgment interframe distance of characteristic point pair in adjacent two frames RGB-D image, preset if it is determined that interframe distance is greater than
Value, then it represents that angle point lazy weight, then ORB characteristic point and description in RGB-D image are calculated, and calculate according to description
The ORB characteristic point pair of adjacent two frames RGB-D image；
It is filtered processing using ORB characteristic point of the RANSAC algorithm to acquisition, following formula is recycled to calculate adjacent two frame
The rotational translation matrix of RGB-D image:
Wherein, n is the number of adjacent two frames RGB-D image characteristic point pair, piWith qiI-th in respectively adjacent two frames RGB-D image
A characteristic point, R, T are respectively rotation, translation matrix；To make E is R, T of minimum value as final result.
2. the method according to claim 1, wherein described be split collected RGB-D image, and counting
The color and shape feature of rear RGB-D image are cut in point counting, are obtained corresponding semantic information and are included:
Super-pixel segmentation is carried out respectively to collected each frame RGB-D image using 2D image segmentation algorithm；
The color and shape feature of each super-pixel are calculated using computer vision algorithms make, recycle support vector machines classification
Device classifies to the color calculated with shape feature, obtains the classifier of several semantic classes；Several languages finally obtained
The classifier of adopted classification is classifier corresponding with color and shape feature classification, semantic classes quantity also with color and shape
Feature classification is corresponding；
Classified using the semantic classes classifier of acquisition to the super-pixel in each frame RGB-D image, to obtain RGB-D
The semantic information of image.
3. method according to claim 1 or 2, which is characterized in that the semantic information by acquisition and interior 3D map
It is merged, obtaining interior 3D semanteme map includes:
Indoor 3D map is divided into several voxels；
Using the Dense CRF model in machine learning and the semantic information of the RGB-D image obtained is combined to carry out voxel semanteme
Information inference determines the semantic information of each voxel, forms final semantic 3D scene map.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510175129.0A CN104732587B (en) | 2015-04-14 | 2015-04-14 | A kind of indoor 3D semanteme map constructing method based on depth transducer |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510175129.0A CN104732587B (en) | 2015-04-14 | 2015-04-14 | A kind of indoor 3D semanteme map constructing method based on depth transducer |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104732587A CN104732587A (en) | 2015-06-24 |
| CN104732587B true CN104732587B (en) | 2019-02-01 |
Family
ID=53456455
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510175129.0A Active CN104732587B (en) | 2015-04-14 | 2015-04-14 | A kind of indoor 3D semanteme map constructing method based on depth transducer |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104732587B (en) |
Families Citing this family (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10019821B2 (en) | 2014-09-02 | 2018-07-10 | Naver Business Platform Corp. | Apparatus and method for constructing indoor map using cloud point |
| CN105045263B (en) * | 2015-07-06 | 2016-05-18 | 杭州南江机器人股份有限公司 | A kind of robot method for self-locating based on Kinect depth camera |
| WO2017079918A1 (en) * | 2015-11-11 | 2017-05-18 | 中国科学院深圳先进技术研究院 | Indoor scene scanning reconstruction method and apparatus |
| TWI639133B (en) | 2015-11-13 | 2018-10-21 | 南韓商納寶商務平台股份有限公司 | Apparatus and method for constructing indoor map using cloud point |
| CN105665970B (en) * | 2016-03-01 | 2018-06-22 | 中国科学院自动化研究所 | For the path point automatic creation system and method for welding robot |
| CN106052674B (en) * | 2016-05-20 | 2019-07-26 | 青岛克路德机器人有限公司 | A kind of SLAM method and system of Indoor Robot |
| CN106067191A (en) * | 2016-05-25 | 2016-11-02 | 深圳市寒武纪智能科技有限公司 | The method and system of semantic map set up by a kind of domestic robot |
| CN106840141A (en) * | 2017-02-02 | 2017-06-13 | 王恒升 | A kind of semantic map of mobile robot indoor navigation |
| CN107063258A (en) * | 2017-03-07 | 2017-08-18 | 重庆邮电大学 | A kind of mobile robot indoor navigation method based on semantic information |
| CN107992036B (en) * | 2017-11-20 | 2021-03-16 | 汕头大学 | Method and device for planning vehicle access path in intelligent parking garage and storage medium |
| WO2019127102A1 (en) * | 2017-12-27 | 2019-07-04 | 深圳前海达闼云端智能科技有限公司 | Information processing method and apparatus, cloud processing device, and computer program product |
| CN108955682A (en) * | 2018-04-03 | 2018-12-07 | 哈尔滨工业大学深圳研究生院 | Mobile phone indoor positioning air navigation aid |
| CN109506658B (en) * | 2018-12-26 | 2021-06-08 | 广州市申迪计算机系统有限公司 | Robot autonomous positioning method and system |
| CN111679661A (en) * | 2019-02-25 | 2020-09-18 | 北京奇虎科技有限公司 | Semantic map construction method based on depth camera and sweeping robot |
| CN110516641B (en) * | 2019-08-30 | 2022-04-12 | 苏州大学 | Construction method of environment map and related device |
| CN112669355B (en) * | 2021-01-05 | 2023-07-25 | 北京信息科技大学 | Method and system for splicing and fusing focusing stack data based on RGB-D super pixel segmentation |
| CN112904437B (en) * | 2021-01-14 | 2023-03-24 | 支付宝(杭州)信息技术有限公司 | Detection method and detection device of hidden component based on privacy protection |
| CN112837372A (en) * | 2021-03-02 | 2021-05-25 | 浙江商汤科技开发有限公司 | Data generation method and device, electronic equipment and storage medium |
| CN113313832B (en) * | 2021-05-26 | 2023-07-04 | Oppo广东移动通信有限公司 | Semantic generation method and device of three-dimensional model, storage medium and electronic equipment |
| CN113989376B (en) * | 2021-12-23 | 2022-04-26 | 贝壳技术有限公司 | Method and device for acquiring indoor depth information and readable storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103712617A (en) * | 2013-12-18 | 2014-04-09 | 北京工业大学 | Visual-content-based method for establishing multi-level semantic map |
| CN104330090A (en) * | 2014-10-23 | 2015-02-04 | 北京化工大学 | Robot distributed type representation intelligent semantic map establishment method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2014146668A2 (en) * | 2013-03-18 | 2014-09-25 | Aalborg Universitet | Method and device for modelling room acoustic based on measured geometrical data |
-
2015
- 2015-04-14 CN CN201510175129.0A patent/CN104732587B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103712617A (en) * | 2013-12-18 | 2014-04-09 | 北京工业大学 | Visual-content-based method for establishing multi-level semantic map |
| CN104330090A (en) * | 2014-10-23 | 2015-02-04 | 北京化工大学 | Robot distributed type representation intelligent semantic map establishment method |
Non-Patent Citations (1)
| Title |
|---|
| "面向服务机器人的室内语义地图构建的研究";王锋;《中国博士学位论文全文数据库 信息科技辑》;20141015;论文摘要，第41-64、103-107页 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104732587A (en) | 2015-06-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104732587B (en) | A kind of indoor 3D semanteme map constructing method based on depth transducer | |
| US11107272B2 (en) | Scalable volumetric 3D reconstruction | |
| US11216971B2 (en) | Three-dimensional bounding box from two-dimensional image and point cloud data | |
| Tatarchenko et al. | Tangent convolutions for dense prediction in 3d | |
| Zhang et al. | Meshstereo: A global stereo model with mesh alignment regularization for view interpolation | |
| CN107622244B (en) | Indoor scene fine analysis method based on depth map | |
| Zhu et al. | RGB-D local implicit function for depth completion of transparent objects | |
| CN111243093B (en) | Three-dimensional face grid generation method, device, equipment and storage medium | |
| Liang et al. | Image based localization in indoor environments | |
| Monica et al. | Contour-based next-best view planning from point cloud segmentation of unknown objects | |
| CN109737974A (en) | A kind of 3D navigational semantic map updating method, device and equipment | |
| Rusu | Semantic 3D object maps for everyday robot manipulation | |
| CN110363817B (en) | Target pose estimation method, electronic device, and medium | |
| Parkison et al. | Semantic Iterative Closest Point through Expectation-Maximization. | |
| CN110648397A (en) | Scene map generation method and device, storage medium and electronic equipment | |
| Bewley et al. | Advantages of exploiting projection structure for segmenting dense 3D point clouds | |
| CN112784873A (en) | Semantic map construction method and equipment | |
| EP4365841A1 (en) | Object pose detection method and apparatus, computer device, and storage medium | |
| CN115439607A (en) | Three-dimensional reconstruction method and device, electronic equipment and storage medium | |
| Tao et al. | Indoor 3D semantic robot VSLAM based on mask regional convolutional neural network | |
| Zhang et al. | Research on 3D architectural scenes construction technology based on augmented reality | |
| CN107330930B (en) | Three-dimensional image depth information extraction method | |
| Wang et al. | Understanding of wheelchair ramp scenes for disabled people with visual impairments | |
| Li et al. | Improving autonomous exploration using reduced approximated generalized voronoi graphs | |
| CN112819937B (en) | Self-adaptive multi-object light field three-dimensional reconstruction method, device and equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |