CN114119940A - Obstacle point cloud convex hull segmentation method based on RANSAC and angular point extraction - Google Patents
Obstacle point cloud convex hull segmentation method based on RANSAC and angular point extraction Download PDFInfo
- Publication number
- CN114119940A CN114119940A CN202111488317.0A CN202111488317A CN114119940A CN 114119940 A CN114119940 A CN 114119940A CN 202111488317 A CN202111488317 A CN 202111488317A CN 114119940 A CN114119940 A CN 114119940A
- Authority
- CN
- China
- Prior art keywords
- point
- point cloud
- angular
- points
- corner
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T19/00—Manipulating 3D models or images for computer graphics
- G06T19/20—Editing of 3D images, e.g. changing shapes or colours, aligning objects or positioning parts
Landscapes
- Engineering & Computer Science (AREA)
- Architecture (AREA)
- Computer Graphics (AREA)
- Computer Hardware Design (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method for partitioning a convex hull of a point cloud of an obstacle based on RANSAC and angular point extraction, which comprises the following steps: performing at least one time of downsampling and dimensionality reduction on the obstacle point cloud to obtain a preprocessed two-dimensional point cloud; performing RANSAC linear point cloud extraction and Euclidean clustering on the two-dimensional point cloud for multiple times, traversing each point cloud obtained by the Euclidean clustering, and performing angular point extraction; determining the gravity center according to the angular points of the two-dimensional point cloud, simultaneously determining polar angles of the angular points, and sequencing the angular points in a clockwise/counterclockwise manner according to the polar angles corresponding to the angular points; traversing the sequenced angular points, judging concave points and issuing a two-dimensional convex hull until all the angular points are traversed; and (4) carrying out convex hull segmentation according to the concave-convex points. The invention provides an obstacle point cloud convex hull segmentation method based on RANSAC and angular point extraction.
Description
Technical Field
The invention relates to the field of robot environment perception, in particular to a method for segmenting a two-dimensional convex hull of an obstacle point cloud based on RANSAC and angular point extraction.
Background
The shape, size and position of the convex hull form of the obstacle need to be known when the robot stops and avoids the obstacle. The general method is that the obstacle point cloud is compressed into two dimensions from three dimensions, then the two dimensions are framed and issued in a two-dimensional convex hull mode, and the robot plans an obstacle avoidance path according to the two-dimensional convex hull information of the obstacle. However, the current point cloud two-dimensional convex hull algorithm is not ideal: due to the shape of the obstacle, the convex hull often wraps the non-obstacle area, resulting in a reduction in the travelable area of the robot.
Disclosure of Invention
An object of the present invention is to solve at least the above problems and/or disadvantages and to provide at least the advantages described hereinafter.
To achieve these objects and other advantages and in accordance with the purpose of the invention, there is provided an obstacle point cloud convex hull segmentation method based on RANSAC and corner extraction, comprising:
firstly, performing down-sampling and dimension reduction processing on the obstacle point cloud at least once to obtain a preprocessed two-dimensional point cloud;
step two, performing RANSAC linear point cloud extraction and Euclidean clustering on the two-dimensional point cloud for multiple times, stopping the linear point cloud extraction until the proportion of the residual points in the initial point cloud is smaller than a preset proportion threshold value, and directly performing Euclidean clustering on the residual point cloud;
step three, traversing each point cloud obtained by European clustering, and extracting angular points;
determining the gravity center according to the angular points of the two-dimensional point cloud, simultaneously determining polar angles of the angular points, and performing clockwise/counterclockwise sorting on the angular points according to the polar angles corresponding to the angular points;
step five, traversing the sequenced angular points, judging concave points and issuing two-dimensional convex hulls until all the angular points are traversed;
and sixthly, performing convex hull segmentation according to the concave-convex points.
Preferably, in the first step, the two-dimensional point cloud is acquired in a manner and configured to include:
s10, transforming the obstacle point cloud from the camera coordinate system to the position below the center of the robot body;
s11, performing primary down-sampling on the obstacle point cloud;
s12, performing dimensionality reduction on the obstacle point cloud;
s13, carrying out secondary down-sampling on the point cloud after the dimensionality reduction;
wherein the dimension reduction process comprises:
for the right-hand system with the z-axis parallel to the world coordinate system, the z-value of the point in the obstacle point cloud becomes 0, and for other coordinate systems, the obstacle point cloud is projected onto the spatial plane of the known model parameters.
Preferably, in the second step, when performing the euclidean clustering on the remaining point clouds, the parameters are set to be consistent with the down-sampling parameters, so as to ensure that the geometric center of the small point cloud is within the point cloud.
Preferably, in step three, the angular point extraction method is as follows:
s30, after down-sampling the two-dimensional point cloud, taking each down-sampled two-dimensional point as a dot, and taking a parameter value during down-sampling as a side length to construct a square for each dot;
s31, determining coordinate information of four vertexes of the square according to the down-sampling parameters;
and S32, traversing the vertex coordinates of all squares, and regarding the vertex with odd occurrence times as a corner point.
5. The method as claimed in claim 1, wherein in step four, the center of gravity of the corner points is determined by calculating the mean value of all the corner points in x coordinate and y coordinate to obtain the center of gravity of the two-dimensional point cloud.
Preferably, in the fifth step, each corner point is traversed according to the sorting order of the polar angles, and when the corner point k is traversed, if the vector is a vector
And
if the cross product between the points is less than 0, the angular point k is a convex point; if vector
And
if the cross product between the two angular points is greater than 0, the angular point k is a concave point, an included angle between the two vectors is calculated, when the included angle is greater than the minimum concave angle threshold value and less than 180 degrees, the angular point is reserved, otherwise, the angular point is rejected.
Preferably, in step five, the minimum reentrant angle threshold is adjusted to control the number of the finally issued two-dimensional convex hulls, that is, the smaller the minimum reentrant angle threshold is, the more the two-dimensional convex hulls are; the larger the minimum reentrant angle threshold, the fewer the two-dimensional convex hulls.
Preferably, in the sixth step, the convex hull division according to the concave and convex points is configured to include:
traversing counterclockwise from the first corner point, if the current corner point is the second corner point and is a convex point, and the cross product of the included angles of two vectors formed by the gravity center and the first corner point and the included angles of two vectors formed by the gravity center and the second corner point is positive, determining that the second corner point is one corner point of the convex hull; if the cross product of the included angles of two vectors formed by the gravity center and the first angular point and the gravity center and the second angular point is negative, the current convex hull traversal should be stopped at the upper angular point, and all the angular points between the initial angular point and the stopping angular point and the gravity center of the point cloud are issued as convex hull angular points in a counterclockwise sequence;
when the angular points traverse to the concave points, the traversal is stopped, and all the angular points and the gravity center between the initial point of the round of traversal and the concave points are issued as convex hull angular points in a counterclockwise sequence.
The invention at least comprises the following beneficial effects: firstly, the two-dimensional convex hull wrapping area is closer to the actual area of the obstacle based on the RANSAC algorithm and the angular point extraction, and the travelable area of the robot is increased.
Secondly, the method is suitable for any sensor with point cloud as sensing data, has strong universality and only needs to be correspondingly adapted to the sensor.
Thirdly, the method of the invention can set a reentrant angle judgment threshold value, reduces the envelope of the non-obstacle area and achieves good envelope effect.
Fourthly, the method is not only suitable for the two-dimensional convex hull segmentation of the point cloud, but also suitable for the convex hull segmentation of a single connected region on a two-dimensional image.
Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention.
Drawings
FIG. 1 is a flow process diagram of the present invention;
FIG. 2 is a schematic diagram of point cloud distribution after RANSAC linear point cloud extraction and Euclidean clustering according to the present invention;
FIG. 3 is a schematic diagram of the corner distribution of a two-dimensional point cloud after down-sampling according to the present invention;
FIG. 4 is a polar angle diagram of the corner point of FIG. 3;
FIG. 5 is a schematic diagram of concave points corresponding to the corner points after the center of gravity of the point cloud is determined according to the present invention;
FIG. 6 is a schematic diagram of salient points corresponding to angular points after determining the gravity centers of the point clouds in the invention;
FIG. 7 is a schematic diagram of convex hull segmentation according to concave and convex points after determining the center of gravity of a point cloud according to the present invention;
FIG. 8 is a diagram illustrating the segmentation effect of base in the prior art;
fig. 9 is a schematic diagram of the improved segmentation effect of the present invention.
Detailed Description
The present invention is further described in detail below with reference to the attached drawings so that those skilled in the art can implement the invention by referring to the description text.
The method enables the wrapping area of the two-dimensional convex hull to be closer to the actual area of the obstacle based on the RANSAC algorithm and the angular point extraction, reduces the reduction of the travelable area of the robot, is suitable for sensors taking sensing data such as a 2D/3D laser radar, an RGB-D depth camera and the like as point clouds, has a better enveloping effect on the obstacle by the two-dimensional convex hull segmented after the obstacle two-dimensional projection, is close to the area of the obstacle two-dimensional projection, and is also suitable for convex hull segmentation of a single-connection area on a two-dimensional image.
Specifically, the present invention comprises the steps of:
1. down-converting the obstacle point cloud from the camera coordinate system/camera _ depth _ optical _ frame to be below the robot body center/base _ FOOTprint: the/base _ footing coordinate system is a commonly used right-hand system, and the z-axis is parallel to the z-axis of the world coordinate system, so that the point cloud processing is convenient.
2. Down-sampling the obstacle point cloud: the number of points of the initial obstacle point cloud is large, and the initial point cloud needs to be subjected to down-sampling in order to accelerate the operation speed; the point cloud down-sampling parameters are determined according to the actual application scene: typically the down-sampling parameters are smaller indoors as compared to outdoors.
3. Reducing the dimension of the obstacle point cloud: the method is to perform convex hull segmentation on two-dimensional point cloud, and the dimension reduction method generally comprises the following steps: a. for a right-hand system with the z axis parallel to the world coordinate system, the purpose of reducing the dimension can be achieved by changing the z value of a point in the obstacle point cloud into 0; b. for other coordinate systems, the cloud of obstacle points is typically projected onto a spatial plane of known model parameters.
4. The point cloud after dimensionality reduction is subjected to down sampling, namely the three-dimensional point cloud is unevenly distributed after dimensionality reduction to two dimensions, and the two-dimensional point cloud needs to be subjected to down sampling again in order to facilitate angular point extraction; the point cloud down-sampling parameters are determined according to the actual application scene: typically the down-sampling parameters are smaller indoors as compared to outdoors.
5. RANSAC linear point cloud extraction and Euclidean clustering: RANSAC straight-line point cloud extraction is performed on the two-dimensional point cloud, as shown by a rectangular dotted line frame in FIG. 2. The parameters of RANSAC straight line extraction can be determined according to the actual scene: usually, the number of straight line extractions indoors is larger than the distance threshold of the outer point outdoors and smaller. Meanwhile, the extracted linear point cloud A needs to be subjected to Euclidean clustering, the linear point cloud with broken lines is separated, the linear point cloud with broken lines is prevented from being enveloped by a convex hull, the feasible range of the robot is reduced, and a rectangular dotted line frame in a rectangular frame B in fig. 2 is the broken line of the linear point cloud. The parameters of the Euclidean clustering are determined according to actual scenes: generally, the Euclidean clustering parameter is larger outdoors than indoors.
6. Repeating the linear point cloud extraction and the Euclidean clustering until the proportion of the residual points in the initial point cloud is less than a proportional threshold, and carrying out the Euclidean clustering on the residual point cloud: and setting a proportion threshold value, and stopping the extraction of the straight-line point clouds when the number of the residual point clouds is less than the initial point clouds multiplied by the proportion threshold value. The remaining point clouds are substantially a small portion of aggregated point clouds, and are directly subjected to euclidean clustering, as shown by the dotted circle in fig. 2. The parameters of the remaining point cloud Euclidean clustering are set to be the same as the down-sampling parameters, and the geometric center of the small point cloud is ensured to be in the point cloud as much as possible.
7. And traversing each point cloud obtained by Euclidean clustering, and extracting angular points: the distribution characteristics of the two-dimensional point cloud after downsampling are shown in fig. 3: the blue round point in the middle of the square is a two-dimensional point C after down sampling; the side length of the square is equal to the size of the parameters during downsampling, and the coordinate information of four vertexes of the square can be determined according to the downsampling parameters; and traversing all the coordinates of the vertexes of the squares, wherein the vertexes with odd times of occurrence are the angular points, as shown by the dots D in FIG. 3.
8. Determining the gravity center according to the angular points of the two-dimensional point cloud, and simultaneously determining the polar angle of each angular point: and solving the mean value of x coordinates and the mean value of y coordinates of all the angular points to obtain the gravity center of the two-dimensional point cloud. Because the point clouds after RANSAC straight line extraction and Euclidean clustering are basically in straight line distribution and small block aggregation distribution, the center of gravity generally falls in the point clouds, and is close to the point clouds under extreme conditions. Taking fig. 3 as an example, the polar angle distribution of the corner points is shown in fig. 4, and the center of gravity of the point cloud is E.
9. And (3) sequencing the angular points according to the polar angles corresponding to the angular points by using a clockwise/anticlockwise pointer: as shown in fig. 4, the corner points are sorted counterclockwise according to the polar angle size.
10. Traversing the sequenced angular points, judging concave points and issuing two-dimensional convex hulls until all the angular points are traversed: as shown in fig. 5, each corner point is traversed in polar-angular order. When traversing to the corner point k, if the vector
And
when the cross product between the two points is less than 0, the angular point k is a convex point; if the vector is as shown in FIG. 6
And
if the cross product between the two vectors is greater than 0, the angular point k is a concave point, and the included angle between the two vectors should be calculated: and when the included angle is larger than the minimum reentrant angle threshold and is smaller than 180 degrees, the angular point is reserved, otherwise, the angular point is rejected. Adjusting the minimum reentrant angle threshold may control the number of two-dimensional convex hulls that are eventually issued: the smaller the minimum reentrant angle threshold, the more two-dimensional convex hulls; the larger the minimum reentrant angle threshold, the fewer the two-dimensional convex hulls.
11. And (3) carrying out convex hull segmentation according to concave and convex points: as shown in fig. 7, counterclockwise traversal is started from a first corner point (corner point 1), if the current corner point is a second corner point (corner point 2) and is a convex point, and the cross product of the included angles of two vectors formed by the gravity center and the first corner point, and the gravity center and the second corner point is positive, the second corner point is determined to be one corner point of the convex hull at this time; and if the cross product of the included angles of two vectors formed by the gravity center and the first angular point and the gravity center and the second angular point is negative, stopping the traversal of the current convex hull at the upper angular point, and releasing all the angular points between the initial angular point and the stopping angular point and the gravity center of the point cloud as convex hull angular points in a counterclockwise sequence. When the angular points traverse to the concave points, the traversal is stopped, and all the angular points and the gravity center between the initial point of the round of traversal and the concave points are issued as convex hull angular points in a counterclockwise sequence.
The method enables the two-dimensional convex hull wrapping area to be closer to the actual area of the obstacle based on the RANSAC algorithm and the corner extraction, and increases the driving area of the robot.
Comparative experiment: the more reasonable passable area can be accurately calculated in real time, and a certain fixed scene of a 5 × 5m map is taken as an example to be compared on an X86 notebook.
1. And (3) comparing the segmentation effects: compared with the prior art, the invention has the advantages that the comparison effect is shown in fig. 8-9, and the barrier convex hull is arranged in the green envelope curve in fig. 8-9, so that the invention can more accurately divide the barrier, reserve more passable areas for the robot, and improve the flexibility of the robot motion.
2. And (3) speed comparison:
| item | Base | Ours |
| Speed (HZ) | 15 | 30 |
| Whether the phenomenon of pause exists when the scene is complex | Is that | Whether or not |
As can be seen from the above table, compared with the prior art, the processing speed of the present invention can be doubled, and the present invention does not have a stuck phenomenon when applied in a complex scene.
The above scheme is merely illustrative of a preferred example, and is not limiting. When the invention is implemented, appropriate replacement and/or modification can be carried out according to the requirements of users.
The number of apparatuses and the scale of the process described herein are intended to simplify the description of the present invention. Applications, modifications and variations of the present invention will be apparent to those skilled in the art.
While embodiments of the invention have been disclosed above, it is not intended to be limited to the uses set forth in the specification and examples. It can be applied to all kinds of fields suitable for the present invention. Additional modifications will readily occur to those skilled in the art. It is therefore intended that the invention not be limited to the exact details and illustrations described and illustrated herein, but fall within the scope of the appended claims and equivalents thereof.
Claims (8)
1. A method for segmenting a convex hull of a point cloud of an obstacle based on RANSAC and angular point extraction is characterized by comprising the following steps:
firstly, performing down-sampling and dimension reduction processing on the obstacle point cloud at least once to obtain a preprocessed two-dimensional point cloud;
step two, performing RANSAC linear point cloud extraction and Euclidean clustering on the two-dimensional point cloud for multiple times, stopping the linear point cloud extraction until the proportion of the residual points in the initial point cloud is smaller than a preset proportion threshold value, and directly performing Euclidean clustering on the residual point cloud;
step three, traversing each point cloud obtained by European clustering, and extracting angular points;
determining the gravity center according to the angular points of the two-dimensional point cloud, simultaneously determining polar angles of the angular points, and performing clockwise/counterclockwise sorting on the angular points according to the polar angles corresponding to the angular points;
step five, traversing the sequenced angular points, judging concave points and issuing two-dimensional convex hulls until all the angular points are traversed;
and sixthly, performing convex hull segmentation according to the concave-convex points.
2. The method for segmentation of convex hull of obstacle point cloud based on RANSAC and corner extraction as claimed in claim 1, wherein in step one, the two-dimensional point cloud is obtained and configured to include:
s10, transforming the obstacle point cloud from the camera coordinate system to the position below the center of the robot body;
s11, performing primary down-sampling on the obstacle point cloud;
s12, performing dimensionality reduction on the obstacle point cloud;
s13, carrying out secondary down-sampling on the point cloud after the dimensionality reduction;
wherein the dimension reduction process comprises:
for the right-hand system with the z-axis parallel to the world coordinate system, the z-value of the point in the obstacle point cloud becomes 0, and for other coordinate systems, the obstacle point cloud is projected onto the spatial plane of the known model parameters.
3. The method as claimed in claim 1, wherein in step two, when performing Euclidean clustering on the remaining point clouds, the parameters are set to be consistent with the down-sampling parameters so as to ensure that the geometric center of the small point cloud is within the point cloud.
4. The method of claim 1, wherein in step three, the corner points are extracted by:
s30, after down-sampling the two-dimensional point cloud, taking each down-sampled two-dimensional point as a dot, and taking a parameter value during down-sampling as a side length to construct a square for each dot;
s31, determining coordinate information of four vertexes of the square according to the down-sampling parameters;
and S32, traversing the vertex coordinates of all squares, and regarding the vertex with odd occurrence times as a corner point.
5. The method as claimed in claim 1, wherein in step four, the center of gravity of the corner points is determined by calculating the mean value of all the corner points in x coordinate and y coordinate to obtain the center of gravity of the two-dimensional point cloud.
6. The method as claimed in claim 1, wherein in step five, each corner point is traversed according to the order of polar angles, and when the corner point k is traversed, if the vector is the same as the vector, the method comprises
And
if the cross product between the points is less than 0, the angular point k is a convex point; if vector
And
if the cross product between the two angular points is greater than 0, the angular point k is a concave point, an included angle between the two vectors is calculated, when the included angle is greater than the minimum concave angle threshold value and less than 180 degrees, the angular point is reserved, otherwise, the angular point is rejected.
7. The method as claimed in claim 1, wherein in step five, the number of two-dimensional convex hulls to be finally issued is controlled by adjusting the minimum reentrant angle threshold, i.e. the smaller the minimum reentrant angle threshold, the more two-dimensional convex hulls; the larger the minimum reentrant angle threshold, the fewer the two-dimensional convex hulls.
8. The method for segmenting the convex hull of the obstacle point cloud based on RANSAC and corner extraction as claimed in claim 1, wherein in the sixth step, the convex hull segmentation according to the concave-convex points is configured to include:
traversing counterclockwise from the first corner point, if the current corner point is the second corner point and is a convex point, and the cross product of the included angles of two vectors formed by the gravity center and the first corner point and the included angles of two vectors formed by the gravity center and the second corner point is positive, determining that the second corner point is one corner point of the convex hull;
if the cross product of the included angles of two vectors formed by the gravity center and the first angular point and the gravity center and the second angular point is negative, stopping the traversal of the current convex hull at the upper angular point, and releasing all the angular points between the initial angular point and the stopping angular point and the gravity center of the point cloud as convex hull angular points in a counterclockwise sequence;
when the angular points traverse to the concave points, the traversal is stopped, and all the angular points and the gravity center between the initial point of the round of traversal and the concave points are issued as convex hull angular points in a counterclockwise sequence.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111488317.0A CN114119940A (en) | 2021-12-07 | 2021-12-07 | Obstacle point cloud convex hull segmentation method based on RANSAC and angular point extraction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111488317.0A CN114119940A (en) | 2021-12-07 | 2021-12-07 | Obstacle point cloud convex hull segmentation method based on RANSAC and angular point extraction |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114119940A true CN114119940A (en) | 2022-03-01 |
Family
ID=80367591
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111488317.0A Pending CN114119940A (en) | 2021-12-07 | 2021-12-07 | Obstacle point cloud convex hull segmentation method based on RANSAC and angular point extraction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114119940A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115793652A (en) * | 2022-11-30 | 2023-03-14 | 上海木蚁机器人科技有限公司 | Driving control method and device and electronic equipment |
| CN115857502A (en) * | 2022-11-30 | 2023-03-28 | 上海木蚁机器人科技有限公司 | Travel control method and electronic device |
| CN116148809A (en) * | 2023-04-04 | 2023-05-23 | 中储粮成都储藏研究院有限公司 | Automatic generation method and system for grain vehicle sampling point based on laser radar scanning and positioning |
| FR3137202A1 (en) * | 2022-06-28 | 2023-12-29 | Continental Autonomous Mobility Germany | Method for detecting a global concavity on a polygonal line |
-
2021
- 2021-12-07 CN CN202111488317.0A patent/CN114119940A/en active Pending
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| FR3137202A1 (en) * | 2022-06-28 | 2023-12-29 | Continental Autonomous Mobility Germany | Method for detecting a global concavity on a polygonal line |
| WO2024003110A1 (en) * | 2022-06-28 | 2024-01-04 | Continental Autonomous Mobility Germany GmbH | Method for detecting the overall concavity of a polygonal line |
| CN115793652A (en) * | 2022-11-30 | 2023-03-14 | 上海木蚁机器人科技有限公司 | Driving control method and device and electronic equipment |
| CN115857502A (en) * | 2022-11-30 | 2023-03-28 | 上海木蚁机器人科技有限公司 | Travel control method and electronic device |
| CN115857502B (en) * | 2022-11-30 | 2023-12-12 | 上海木蚁机器人科技有限公司 | Driving control method and electronic device |
| CN116148809A (en) * | 2023-04-04 | 2023-05-23 | 中储粮成都储藏研究院有限公司 | Automatic generation method and system for grain vehicle sampling point based on laser radar scanning and positioning |
| CN116148809B (en) * | 2023-04-04 | 2023-06-20 | 中储粮成都储藏研究院有限公司 | Automatic generation method and system for grain vehicle sampling point based on laser radar scanning and positioning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114119940A (en) | Obstacle point cloud convex hull segmentation method based on RANSAC and angular point extraction | |
| CN110826357B (en) | Method, device, medium and equipment for three-dimensional detection and intelligent driving control of object | |
| US9630320B1 (en) | Detection and reconstruction of an environment to facilitate robotic interaction with the environment | |
| EP3689215B1 (en) | Region attribute determination | |
| CN111598916A (en) | Preparation method of indoor occupancy grid map based on RGB-D information | |
| CN112070838B (en) | Object identification and positioning method and device based on two-dimensional-three-dimensional fusion characteristics | |
| WO2021016751A1 (en) | Method for extracting point cloud feature points, point cloud sensing system, and mobile platform | |
| CN113345008B (en) | Laser radar dynamic obstacle detection method considering wheel type robot position and posture estimation | |
| CN111141264B (en) | Unmanned aerial vehicle-based urban three-dimensional mapping method and system | |
| CN104463108A (en) | Monocular real-time target recognition and pose measurement method | |
| CN112541097A (en) | Method and computing system for object identification | |
| Litomisky et al. | Removing moving objects from point cloud scenes | |
| CN110567441B (en) | Particle filter-based positioning method, positioning device, mapping and positioning method | |
| CN109948413A (en) | Method for detecting lane lines based on the fusion of high-precision map | |
| CN110097598B (en) | Three-dimensional object pose estimation method based on PVFH (geometric spatial gradient frequency) features | |
| CN115546202B (en) | Tray detection and positioning method for unmanned forklift | |
| CN111998862B (en) | BNN-based dense binocular SLAM method | |
| CN107680168A (en) | Lattice simplified method based on plane fitting in three-dimensional reconstruction | |
| CN113640826A (en) | Obstacle identification method and system based on 3D laser point cloud | |
| CN114764885A (en) | Obstacle detection method and device, computer-readable storage medium and processor | |
| CN107797556B (en) | A method of realizing server start and stop using Xun Wei robots | |
| CN107122782B (en) | Balanced semi-dense stereo matching method | |
| CN115170648B (en) | Carriage pose determining method and device | |
| CN110264481B (en) | Box-like point cloud segmentation method and device | |
| WO2021131990A1 (en) | Information processing device, information processing method, and program |
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 |