CN104616345B - Octree forest compression based three-dimensional voxel access method - Google Patents
Octree forest compression based three-dimensional voxel access method Download PDFInfo
- Publication number
- CN104616345B CN104616345B CN201410765462.2A CN201410765462A CN104616345B CN 104616345 B CN104616345 B CN 104616345B CN 201410765462 A CN201410765462 A CN 201410765462A CN 104616345 B CN104616345 B CN 104616345B
- Authority
- CN
- China
- Prior art keywords
- block
- voxel
- octree
- node
- dimensional
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
- G06T17/005—Tree description, e.g. octree, quadtree
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Computer Graphics (AREA)
- Geometry (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Processing Or Creating Images (AREA)
- Image Generation (AREA)
Abstract
The invention discloses an octree forest compression based three-dimensional voxel access method. The method comprises the steps of (1) uniformly dividing a voxel space according to the thickness, and creating an octree in each area; (2) performing dynamic construction and growth update for an octree forest according to each frame depth map and camera parameters; (3) quickly searching leaf data by the layered search method according to the constructed forest structure. According to the method, the octree forest sturctuer is utilized to effectively compress the consumption of a storage resource of space voxel data, and meanwhile, the data searching speed and the surface reconstruction precision can be ensured.
Description
Technical field
The invention belongs to three-dimensional reconstruction field, and in particular to a kind of three-dimensional voxel based on the compression of Octree forest is deposited
Take method.
Background technology
Three-dimensional reconstruction is a kind of computer technology for recovering object dimensional information (shape etc.) using two-dimensional projection.Traditional
Three-dimensional surface rebuilding is that object is shot in different visual angles using video camera, the angle point in abstract image, edge, texture,
The essential characteristics such as lines, border, object general profile is recovered according to visual angle relation.Recently as the hair of depth measuring devices
Exhibition, people start by laser measuring apparatus, and ToF (when m- light) equipment such as depth camera and structure light depth camera is entered
The more accurate denser resurfacing of row.With developing rapidly for computer software and hardware, extensive, high-precision three-dimensional scene is painted
Real-time processed is improved constantly, and the difficulty of three-dimensional reconstruction is greatly reduced, and precision is higher, and speed is faster.Three-dimensional reconstruction has wide
General is applied to the fields such as reverse-engineering, video display amusement, industrial design and historical relic's protection.In satellite exploration landforms field, profit
With the great amount of images information taken by satellite, three-dimensional reconstruction is carried out to geological surface, so that more geomorphology informations are obtained, and then
Help is provided to related science research；In medical domain, by the surface model of three-dimensional reconstruction human internal organs, Neng Goubang
Doctor is helped to be more effectively carried out illness analysis and diagnosis；In historical relic's protection field, the historical relic to laying special stress on protecting carries out three-dimensional reconstruction,
The information data of historical relic is preserved in time so that heritage information has diversity and continuity.
In the end of the year 2011, Microsoft Research proposes the reality as collecting device with the depth camera of consumer level (Kinect)
When dense resurfacing and tracing algorithm (KinectFusion).The core of KinectFusion algorithms is to block distance function value
(TSDF), it is one individual first defined in actual physics space, then this individuality is divided into many bodies of uniform size
Element, wherein each voxel store a TSDF value to represent the voxel to the distance on nearest surface, i.e., apart from solid object surface
Nearer voxel, its TSDF value is smaller.
Due to the number of voxels that the impartial division methods that number of voxel is larger and uses cause in KinectFusion algorithms
According to needing to take larger memory space, which limit its application under different platform.According to analysis, in TSDF bodies only
It is effective near the number of regions of body surface, and the ratio shared by surface in physical space is very small, therefore TSDF
Volume data has substantial amounts of redundancy, it is possible to use compression method compresses to it, reduces the occupancy of storage resource.
The content of the invention
For the above-mentioned technical problem existing for prior art, the invention provides a kind of based on the compression of Octree forest
Three-dimensional voxel access method, can be effective to reduce accounting for for storage resource on the premise of former algorithm reconstruction accuracy is ensured
Consumption, realizes the efficient access of TSDF data.
A kind of three-dimensional voxel access method based on the compression of Octree forest is as follows：
Operated for storage：First, rough even partition is carried out to three-dimensional space, obtains multiple identical squares
Block；Then, Octree is built in described block, i.e., dynamic is built since root node using depth camera back projection
Setting and corresponding to carries out block segmentation；Finally, by TSDF data storages in the voxel corresponding to Octree bottom leaf node；
For read operation：First, according to access is treated according to the coordinate in three-dimensional space, the area belonging to it is determined
Block；Then, the block Octree is scanned for using the searching method of depth-first, with search tree in treat access according to seat
The bottom leaf node that mark matches；Finally, corresponding TSDF data are read from the corresponding voxel of the leaf node for searching.
The specific method of described structure Octree is：By block to should be used as the root node of Octree, if by judging
Block need to be divided, then block is evenly divided into eight voxels of identical square, eight voxels correspond to root section respectively
Eight child nodes under point；Similarly, if by judging to divide voxel, voxel further is evenly divided into eight
Small voxel, eight small voxels correspond to eight child nodes under big voxel node respectively；When the Octree number of plies reach it is default most
The big number of plies, even if voxel meets the requirement for further dividing, does not also split again to it.
Judge whether block or voxel need the conditioning process for being divided as follows：
First, the central point C of block or voxel is mapped in the projection plane of depth camera by coordinate transform, is obtained
To corresponding depth image vegetarian refreshments U；
Then, depth image vegetarian refreshments U is returned in three-dimensional space according to its depth value back projection, obtains central point C and exist
Subpoint P in space on body surface；
Finally, the three-dimensional coordinate according to central point C and subpoint P judges whether to carry out block or voxel by following
Divide：
If block or voxel do not exist with body surface in three-dimensional space intersecting, following relation is met then to block
Or voxel is divided：
||p-c||≤T
If block or voxel exist with body surface in three-dimensional space intersecting, meet following relation then to block or
Voxel is divided：
Wherein：P and c are respectively the coordinate vector of subpoint P and central point C in three-dimensional space, and l is block or body
The length of side of element, T and ε is the intercept parameter of setting.
If block does not make further division, its corresponding Octree is set for empty, does not have any TSDF numbers in the block
According to.
The expression formula of described coordinate vector p is as follows：
P=R π (K, u, Z (u))+t
Wherein：U is coordinate vectors of the depth image vegetarian refreshments U in three-dimensional space, and K is the internal reference square of depth camera
Battle array, Z (u) is the depth value of depth image vegetarian refreshments U, and R and t is respectively the spin matrix of depth camera external parameter and is translated towards
Amount；π (K, u, Z (u)) is that depth image vegetarian refreshments U is changed to the coordinate under depth camera coordinate system according to its depth value Z (u)
Vector.
The expression formula of described coordinate vector u is as follows：
U=KR-1(c-t)
It is determined that treating that access method of block belonging to is：Calculated by following formula treat access according to belonging to block position to
Amount r, can be found by the arrangement sequence number in position vector r and treat the access specific position of block in three-dimensional space belonging to
Put；
Wherein：X, y, z corresponds to treat access according to the coordinate value on three-dimensional space X-axis, Y-axis, Z axis, and L is block
The length of side,It is downward bracket function, Numx、Numy、NumzCorrespond to treat access according to affiliated block in three-dimensional space X-axis side
To, the arrangement sequence number in Y direction, Z-direction.
When being scanned for block Octree, first determine whether whether the tree is empty tree：Set if it is empty, then show the tree block
In do not have any TSDF data, directly skip over the block that the tree block advances to next tree and scan for；Set if not empty,
The tree block is scanned for using the searching method of depth-first then, to find what is matched with data coordinates to be taken in the tree
Bottom leaf node simultaneously reads the TSDF data in its correspondence voxel；In search procedure, if the non-bottom leaf segment of present node
Point or present node correspondence voxel in TSDF data it is undesirable, then skip over this node, advance to next brother of node after
Continuous search is until search satisfactory TSDF data.
Compared with the prior art, the present invention has following Advantageous Effects：
(1) present invention tool has carried out tree construction pressure to the voxel data being evenly dividing in original KinectFusion algorithms
Contracting, significantly reduces the storage resource consumption of spatial voxel data, and compression efficiency reaches 90%.
(2) present invention tool devises the data search method based on Octree forest structure, and institute's data storage can be entered
Row efficiently quickly reads.
(3) forest structure of present invention tool design causes that the depth of every Octree is smaller, and search speed is faster.
Brief description of the drawings
Fig. 1 is the structural representation in body space.
Fig. 2 (a) and Fig. 2 (b) are respectively node needs two kinds of situation schematic diagrams of subdivision.
Fig. 3 is the structural representation in Octree and its node correspondence space.
Fig. 4 is the storage resource comparison diagram of Octree forest compression method of the present invention and KinectFusion algorithms.
Fig. 5 (a) and Fig. 5 (b) are respectively KinectFusion algorithms surface reconstructed results and its partial enlarged drawing.
Fig. 6 (a) and Fig. 6 (b) are respectively the present invention by the resurfacing result after 4 layers of Octree forest compression and its office
Portion's enlarged drawing.
Fig. 7 (a) and Fig. 7 (b) are respectively the present invention by the resurfacing result after 5 layers of Octree forest compression and its office
Portion's enlarged drawing.
Specific embodiment
In order to more specifically describe the present invention, below in conjunction with the accompanying drawings and specific embodiment is to technical scheme
It is described in detail.
The present embodiment will carry out affiliated Octree forest structure with reference to accompanying drawing to the body space of a 3m × 3m × 3m sizes
Compression.
As shown in figure 1, first by whole individual space carry out it is rough be evenly dividing 30 × 30 × 30 small cubes, each
Cube size is 0.1m, and an Octree, all of Octree structure in such body space are then built in each zonule
Into an Octree forest.One piece of cubical area in each Octree correspondence physical space, and each tree carries
One is set label to mark whether the tree is non-NULL tree.
Specific forest builds as follows with growing method：
When a given amplitude deepness image and its corresponding camera parameters, can judge every according to three-dimensional mapping relations
Tree each node whether need growth segment.The method of judgement is as follows：
Assuming that the central point c=(X, Y, Z) of Octree node correspondence area of spaceT, its region cube length of side is l,
By coordinate transform, point c can be mapped in the projection plane of video camera and find corresponding depth image vegetarian refreshments u=KR-1(c-
T), wherein K is video camera internal reference matrix, and R and t is respectively the spin matrix and translation vector of video camera external parameter.
Then by the pixel according to its back projection that fathoms make the return trip empty between in：
P=R π (K, u, Z (u))+t
Wherein：π (K, u, Z (u)) functional expressions change to camera coordinate system pixel u according to its depth value Z (u)
3D positions.When occur following two situations any one when, then the node region needs further growth to segment：
1. when p points are in the encirclement ball of the node space, such as shown in Fig. 2 (a).The situation be generally present in node space compared with
Greatly, and space includes surface data.Wherein ε asks expansion micro (being typically set to 0.01m) for encirclement；
2. pixel projection point p is less than or equal to block apart from T (generally 0.03m) apart from the distance of node center c, such as Fig. 2
Shown in (b).The situation is generally present in that node space is smaller, and node center is away from the nearer situation in surface；
||p-c||≤T
When node reaches the depth capacity D of OctreemaxWhen (generally 4 or 5), then no longer it is finely divided, and by TSDF
Data are stored in depth capacity node.Therefore the leaf node for depth capacity being located in Octree is the minimum of spatial surface data
Area unit, corresponding to the voxel in former KinectFusion algorithms.The as TSDF data of leaf node storage.
When a new two field picture is obtained, by the method, all nodes in traversal forest, from root node to leaf section
Point is successively grown renewal, while the TSDF data in all of leaf node are updated, the Octree topological structure for building
And its corresponding space structure of node is as shown in Figure 3.
As shown in figure 4, we have chosen the storage resource of one section of 300 cycle tests of frame to being consumed recorded,
After Octree forest compresses, storage resource occupancy is greatly reduced.In the former constant occupancy 512M of KinectFusion algorithms
Deposit, and the maximum occupancy after being compressed under same volume resolution ratio also only has 52M internal memories.
Data search；Required to look up when light is projected the data of leaf node, present embodiment is using a kind of point
The method of layer search is quickly searched, and detailed process is as follows：
Step one：For any point g=(X, Y, Z) in spaceT, according to each tree root node correspondence being evenly dividing
The length of side l in space obtains which tree of the point in forest：
Wherein：To round downwards.
Step 2：Tree label according to tree r judges whether the tree is empty tree, if sky, then illustrate the root vertex
Corresponding region does not exist surface data, can directly skip over the region that the tree region advances to next tree；If the tree is not sky,
Then carrying out step 3 carries out finer search.
Step 3：When tree for non-NULL when, then carry out the searching method of depth-first, according to the Octree built up search for
Leaf node, reads corresponding TSDF data.If in search procedure, certain node does not have leaf node or TSDF data not to be inconsistent
Close and require, then skip over this node, advance to next brother of node and continue search for until searching satisfactory leaf data.
As shown in Fig. 5~7, Kinfu512 is the surface model that original KinectFusion light is rendered；Oct_f4 is 4
The reconstructed results of layer Octree forest, theoretical resolution is identical with Kinfu512；Oct_f5 is 5 layers of reconstruction knot of Octree forest
Really, theoretical resolution is the twice of Kinfu512.It can be seen that reconstructed results and former algorithm after Octree forest compresses
Quite, there is not data degradation, and can reach reconstruction precision higher in result.
The above-mentioned description to embodiment is to be understood that and apply this hair for ease of those skilled in the art
It is bright.Person skilled in the art obviously can easily make various modifications to above-described embodiment, and described herein
General Principle is applied in other embodiment without by performing creative labour.Therefore, the invention is not restricted to above-described embodiment,
Those skilled in the art's announcement of the invention, the improvement made for the present invention and modification all should be in protections of the invention
Within the scope of.
Claims (6)
1. a kind of three-dimensional voxel access method based on the compression of Octree forest, as follows：
Operated for storage：First, rough even partition is carried out to three-dimensional space, obtains the area of multiple identical squares
Block；Then, Octree is built in described block, i.e., dynamic tree is simultaneously since root node using depth camera back projection
Correspondence carries out block segmentation；Finally, by TSDF data storages in the voxel corresponding to Octree bottom leaf node；
Build Octree specific method be：By block to should be used as the root node of Octree, if need to be entered to block by judgement
Row is divided, then block is evenly divided into eight voxels of identical square, and eight voxels correspond to eight under root node respectively
Child node；Similarly, if by judging to divide voxel, voxel further is evenly divided into eight small voxels,
Eight small voxels correspond to eight child nodes under big voxel node respectively；When the Octree number of plies reaches the default maximum number of plies, i.e.,
Voxel is met the requirement for further dividing, also it is not split again；Wherein judge whether block or voxel need to be drawn
The conditioning process for dividing is as follows：
First, the central point C of block or voxel is mapped in the projection plane of depth camera by coordinate transform, obtains right
The depth image vegetarian refreshments U for answering；
Then, depth image vegetarian refreshments U is returned in three-dimensional space according to its depth value back projection, obtains central point C in space
Subpoint P on middle body surface；
Finally, the three-dimensional coordinate according to central point C and subpoint P judges whether to draw block or voxel by following
Point：
If block or voxel do not exist with body surface in three-dimensional space intersecting, following relation is met then to block or body
Element is divided：
||p-c||≤T
If block or voxel exist with body surface in three-dimensional space intersecting, following relation is met then to block or voxel
Divided：
Wherein：P and c are respectively the coordinate vector of subpoint P and central point C in three-dimensional space, and l is block or voxel
The length of side, T and ε are the intercept parameter of setting；
For read operation：First, according to access is treated according to the coordinate in three-dimensional space, the block belonging to it is determined；So
Afterwards, the block Octree is scanned for using the searching method of depth-first, with search tree in data coordinates phase to be taken
The bottom leaf node of matching；Finally, corresponding TSDF data are read from the corresponding voxel of the leaf node for searching.
2. three-dimensional voxel access method according to claim 1, it is characterised in that：If block does not make further division,
Its corresponding Octree is set for empty, does not have any TSDF data in the block.
3. three-dimensional voxel access method according to claim 1, it is characterised in that：The expression formula of described coordinate vector p
It is as follows：
P=R π (K, u, Z (u))+t
Wherein：U is coordinate vectors of the depth image vegetarian refreshments U in three-dimensional space, and K is the internal reference matrix of depth camera, Z
U () is the depth value of depth image vegetarian refreshments U, R and t is respectively the spin matrix and translation vector of depth camera external parameter；π
(K, u, Z (u)) be by depth image vegetarian refreshments U according to its depth value Z (u) change to the coordinate under depth camera coordinate system to
Amount.
4. three-dimensional voxel access method according to claim 3, it is characterised in that：The expression formula of described coordinate vector u
It is as follows：
U=KR-1(c-t)。
5. three-dimensional voxel access method according to claim 1, it is characterised in that：It is determined that treating the side of access block belonging to
Method is：The position vector r for treating access block belonging to is calculated by following formula, is by the arrangement sequence number in position vector r
Can find and treat the access particular location of block in three-dimensional space belonging to；
Wherein：X, y, z corresponds to treat access according to the coordinate value on three-dimensional space X-axis, Y-axis, Z axis, and L is the side of block
It is long,It is downward bracket function, Numx、Numy、NumzCorrespond to treat access according to affiliated block in three-dimensional space X-axis side
To, the arrangement sequence number in Y direction, Z-direction.
6. three-dimensional voxel access method according to claim 1, it is characterised in that：When being scanned for block Octree,
First determine whether whether the tree is empty tree：Set if it is empty, then show there are not any TSDF data in the tree block, directly skip over this
The block that tree block advances to next tree is scanned for；Set if not empty, then using the searching method of depth-first to the tree
Block is scanned for, with find in the tree with the bottom leaf node that data coordinates to be taken match and read its corresponding voxel
TSDF data；In search procedure, if the TSDF numbers in the non-bottom leaf node of present node or present node correspondence voxel
According to undesirable, then this node is skipped over, advance to next brother of node and continue search for until searching satisfactory TSDF
Data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410765462.2A CN104616345B (en) | 2014-12-12 | 2014-12-12 | Octree forest compression based three-dimensional voxel access method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410765462.2A CN104616345B (en) | 2014-12-12 | 2014-12-12 | Octree forest compression based three-dimensional voxel access method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104616345A CN104616345A (en) | 2015-05-13 |
| CN104616345B true CN104616345B (en) | 2017-05-24 |
Family
ID=53150777
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410765462.2A Active CN104616345B (en) | 2014-12-12 | 2014-12-12 | Octree forest compression based three-dimensional voxel access method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104616345B (en) |
Families Citing this family (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106548507A (en) * | 2015-09-16 | 2017-03-29 | 富士通株式会社 | The method and apparatus of three-dimensional reconstruction object |
| CN106558076B (en) * | 2015-09-16 | 2019-06-18 | 富士通株式会社 | The method and apparatus of three-dimensional reconstruction object |
| CN106548466B (en) * | 2015-09-16 | 2019-03-29 | 富士通株式会社 | The method and apparatus of three-dimensional reconstruction object |
| CN107507276B (en) * | 2016-06-14 | 2020-06-23 | 中国石油化工股份有限公司 | Three-dimensional digital core storage method for core slice display in any direction |
| CN109643457B (en) | 2016-10-12 | 2023-04-21 | 惠普发展公司，有限责任合伙企业 | Serializing representations of three-dimensional objects |
| CN108090404A (en) * | 2016-11-23 | 2018-05-29 | 法乐第(北京)网络科技有限公司 | Three dimensional image processing method |
| CN108268879B (en) * | 2016-12-30 | 2020-10-02 | 法法汽车(中国)有限公司 | Three-dimensional full-convolution network implementation method and equipment and electronic equipment |
| CN108805261B (en) | 2017-04-28 | 2021-11-12 | 微软技术许可有限责任公司 | Convolutional neural network based on octree |
| EP3429209A1 (en) * | 2017-07-13 | 2019-01-16 | Thomson Licensing | A method and apparatus for encoding/decoding a point cloud representing a 3d object |
| CN108010126A (en) * | 2017-12-11 | 2018-05-08 | 苏州蜗牛数字科技股份有限公司 | Method and system based on voxel structure large-scale complex landform |
| CN108196266B (en) * | 2017-12-29 | 2021-07-02 | 中山大学 | Vegetation canopy three-dimensional radiation transmission simulation method based on Lidar data |
| CN108921774A (en) * | 2018-06-08 | 2018-11-30 | 广州虎牙信息科技有限公司 | The storage organization and related edit method, apparatus, equipment and storage medium of model |
| WO2020034086A1 (en) * | 2018-08-14 | 2020-02-20 | 深圳前海达闼云端智能科技有限公司 | Three-dimensional reconstruction method and apparatus for scene, and electronic device and storage medium |
| CN109861965B (en) * | 2018-12-05 | 2021-06-18 | 苏州蜗牛数字科技股份有限公司 | Voxel terrain data compression method and system |
| CN112372632A (en) * | 2020-10-12 | 2021-02-19 | 达闼机器人有限公司 | Mechanical arm motion control method and device and computer readable storage medium |
| CN112734902A (en) * | 2020-12-24 | 2021-04-30 | 清华大学天津高端装备研究院 | Method for rapid modeling by using depth camera |
| CN113487690B (en) * | 2021-09-08 | 2021-11-19 | 北京医百科技有限公司 | Depth map processing method, system and storage medium |
| CN113713381B (en) * | 2021-09-09 | 2023-06-20 | 腾讯科技（深圳）有限公司 | Object management method, device, equipment, storage medium and system |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1632834A (en) * | 2003-12-22 | 2005-06-29 | 中国科学院自动化研究所 | Point reconstruction based very large-scale medical image 3D visualization method |
| CN101763652A (en) * | 2009-06-03 | 2010-06-30 | 中国科学院自动化研究所 | Three-dimensional framework fast extraction method based on branch feathers |
| EP2339481A1 (en) * | 2009-12-03 | 2011-06-29 | National Digital Research Centre | Enablement of three-dimensional hosting, indexing, analysing and querying structure for spatial systems |
| CN103236058A (en) * | 2013-04-25 | 2013-08-07 | 内蒙古科技大学 | Method for obtaining volume of interest of four-dimensional heart image |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6791549B2 (en) * | 2001-12-21 | 2004-09-14 | Vrcontext S.A. | Systems and methods for simulating frames of complex virtual environments |
-
2014
- 2014-12-12 CN CN201410765462.2A patent/CN104616345B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1632834A (en) * | 2003-12-22 | 2005-06-29 | 中国科学院自动化研究所 | Point reconstruction based very large-scale medical image 3D visualization method |
| CN101763652A (en) * | 2009-06-03 | 2010-06-30 | 中国科学院自动化研究所 | Three-dimensional framework fast extraction method based on branch feathers |
| EP2339481A1 (en) * | 2009-12-03 | 2011-06-29 | National Digital Research Centre | Enablement of three-dimensional hosting, indexing, analysing and querying structure for spatial systems |
| CN103236058A (en) * | 2013-04-25 | 2013-08-07 | 内蒙古科技大学 | Method for obtaining volume of interest of four-dimensional heart image |
Non-Patent Citations (4)
| Title |
|---|
| Simple and effi cient t raversal methods for quadt rees and oct rees;FRISKEN S 等;《Graphics Tools》;20031231;第7卷(第3期);1-11 * |
| 三维数字矿床与隐伏矿体立体定量预测研究;毛先成;《中国博士学位论文全文数据库》;20061215;第2.4.2节 * |
| 基于八叉树的三维网格模型体素化方法;吴晓军 等;《工程图学学报》;20051231(第4期);1-7 * |
| 基于投影体积与八叉树的三维网格模型体素化方法;穆斌 等;《地理与地理信息科学》;20100731;第26卷(第4期);25-31 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104616345A (en) | 2015-05-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104616345B (en) | Octree forest compression based three-dimensional voxel access method | |
| Wang et al. | Lidar point clouds to 3-D urban models $: $ A review | |
| CN105378796B (en) | Scalable volume 3D reconstruct | |
| Zeng et al. | Octree-based fusion for realtime 3D reconstruction | |
| Zhang et al. | Interactive collision detection for deformable models using streaming AABBs | |
| CN106803267A (en) | Indoor scene three-dimensional rebuilding method based on Kinect | |
| CN108198145A (en) | For the method and apparatus of point cloud data reparation | |
| CN110490917A (en) | Three-dimensional rebuilding method and device | |
| CN102077243A (en) | 3-d texture super resolution image generation device and image generation method | |
| Qin et al. | Aura 3D textures | |
| Scheiblauer | Interactions with gigantic point clouds | |
| CN112530005B (en) | Three-dimensional model linear structure recognition and automatic restoration method | |
| CN111028335B (en) | Point cloud data block surface patch reconstruction method based on deep learning | |
| CN103530907A (en) | Complicated three-dimensional model drawing method based on images | |
| CN117280387A (en) | Displacement micro-grid for ray and path tracing | |
| CN105825471A (en) | Unity-3D-based three-dimensional surface reconstruction and rendering method | |
| Condorelli et al. | A comparison between 3D reconstruction using nerf neural networks and mvs algorithms on cultural heritage images | |
| CN114494611B (en) | Intelligent three-dimensional reconstruction method, device, equipment and medium based on nerve basis function | |
| CN107590858A (en) | Medical sample methods of exhibiting and computer equipment, storage medium based on AR technologies | |
| Kang et al. | Efficient randomized hierarchy construction for interactive visualization of large scale point clouds | |
| CN117315153A (en) | Human body reconstruction and rendering method and device for cooperative light field and occupied field | |
| Cheng et al. | Fast visualization of 3D massive data based on improved Hilbert R-tree and stacked LSTM models | |
| Sahebdivani et al. | Deep learning based classification of color point cloud for 3D reconstruction of interior elements of buildings | |
| Koca et al. | A hybrid representation for modeling, interactive editing, and real-time visualization of terrains with volumetric features | |
| Imielinska et al. | Technical challenges of 3D visualization of large color data sets |
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 |