CN112731339A - Three-dimensional target detection system based on laser point cloud and detection method thereof - Google Patents
Three-dimensional target detection system based on laser point cloud and detection method thereof Download PDFInfo
- Publication number
- CN112731339A CN112731339A CN202110004308.3A CN202110004308A CN112731339A CN 112731339 A CN112731339 A CN 112731339A CN 202110004308 A CN202110004308 A CN 202110004308A CN 112731339 A CN112731339 A CN 112731339A
- Authority
- CN
- China
- Prior art keywords
- point cloud
- module
- cloud data
- dimensional target
- voxelized
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 138
- 238000000605 extraction Methods 0.000 claims abstract description 34
- 230000004927 fusion Effects 0.000 claims abstract description 33
- 238000010586 diagram Methods 0.000 claims description 34
- 239000000284 extract Substances 0.000 claims description 8
- 230000005484 gravity Effects 0.000 claims description 7
- 238000006243 chemical reaction Methods 0.000 claims description 4
- 238000013527 convolutional neural network Methods 0.000 claims description 4
- 238000004220 aggregation Methods 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/48—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S17/00
- G01S7/4802—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S17/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Length Measuring Devices By Optical Means (AREA)
- Optical Radar Systems And Details Thereof (AREA)
Abstract
A three-dimensional target detection system based on laser point cloud comprises a voxel size division module, a feature coding module, a feature extraction and fusion module, a target regression and detection module and a laser radar, wherein the output end of the laser radar is connected with the input end of the target regression and detection module through the voxel size division module, the feature coding module and the feature extraction and fusion module in sequence, firstly, the voxel size division module adopts different voxel scales to perform voxel division on three-dimensional target point clouds obtained from a laser radar to obtain a plurality of voxelized point clouds, then the characteristic coding module performs characteristic coding on the voxelized point clouds, and finally, the target regression and detection module obtains a three-dimensional target detection frame according to the final feature map. The design can ensure that the structural characteristics of the point cloud are not lost, and the detection precision of the three-dimensional target is improved.
Description
Technical Field
The invention belongs to the technical field of target detection, and particularly relates to a three-dimensional target detection system based on laser point cloud and a detection method thereof, which are suitable for improving the detection precision of a three-dimensional target.
Background
With the wide application of laser radars in unmanned driving, a point cloud-based three-dimensional target detection method becomes the mainstream research direction in the future, on one hand, the existing point cloud-based three-dimensional target detection method generally only adopts single voxel size division during point cloud voxelization, and a single voxel size setting is easy to lose some local structures of point cloud data, so that all available information in a detection scene is not enough to be represented, and the detection precision is reduced, on the other hand, the position relation between points is not considered during feature coding of the voxelized point cloud data, so that the feature representation of the point cloud data is not rich, the feature learning is not facilitated, and the detection precision is further reduced.
Chinese patent: the invention of application publication No. CN111199206A and application publication No. 2020.05.26 discloses a three-dimensional target detection method, device, computer equipment and storage medium, firstly, a three-dimensional target detection model is adopted to detect point clouds to be detected, a plurality of detection frames are obtained, then, the plurality of detection frames are expanded, coordinates of laser points in each expanded detection frame are converted to a local coordinate system with the central point of the detection frame where the laser points are located as the origin, point clouds to be detected under the local coordinate system are obtained, then, the point clouds to be detected under the local coordinate system are mapped to a voxel space, voxel characteristics are generated according to the coordinates of the point clouds to be detected under the local coordinate system and the voxel space position coordinates of the point clouds to be detected, then, a sparse convolution network is adopted to carry out characteristic extraction on the voxel characteristics, target characteristics are obtained, finally, a regional candidate network is adopted to detect the target characteristics, and a detection, although the point cloud to be detected is mapped to the voxel space to generate features, and then the sparse convolution network is adopted to extract the features of the voxel to obtain the target features, the extraction of the target features is facilitated, but only single voxel size division is adopted during point cloud voxelization, so that the local structure of point cloud data is easily lost, all available information in a scene is not enough represented, and the problem of low detection precision still exists.
Disclosure of Invention
The invention aims to overcome the problem of low detection precision and provides a laser point cloud-based three-dimensional target detection system with high detection precision and a detection method thereof.
In order to achieve the above purpose, the invention provides the following technical scheme:
a three-dimensional target detection system based on laser point cloud comprises a voxel size division module, a feature coding module, a feature extraction and fusion module and a target regression and detection module, wherein the signal output end of the voxel size division module is connected with the signal input end of the target regression and detection module sequentially through the feature coding module and the feature extraction and fusion module;
the voxel size division module is used for obtaining point cloud data of a three-dimensional target, performing voxel division on the point cloud data by adopting different voxel scales to obtain a plurality of voxelized point cloud data, and then sending the voxelized point cloud data to the feature coding module;
the characteristic coding module is used for firstly carrying out characteristic coding on the voxelized point cloud data and then sending the coded voxelized point cloud data to the characteristic extraction and fusion module;
the characteristic extraction and fusion module is used for sequentially carrying out characteristic extraction and weighted fusion on the encoded voxelized point cloud data to obtain a final characteristic map, and then sending the final characteristic map to the target regression and detection module;
and the target regression and detection module is used for obtaining a three-dimensional target detection frame according to the final characteristic diagram.
The detection system further comprises a laser radar, and the signal input end of the voxel size dividing module is connected with the signal output end of the laser radar.
A detection method of a three-dimensional target detection system based on laser point cloud sequentially comprises the following steps:
firstly, acquiring point cloud data of a three-dimensional target from a laser radar by the voxel size dividing module, dividing the point cloud data by adopting different voxel scales to obtain a plurality of voxelized point cloud data, and then sending the voxelized point cloud data to the feature coding module;
secondly, the characteristic coding module carries out characteristic coding on the voxelized point cloud data, and then sends the coded voxelized point cloud data to the characteristic extraction and fusion module;
thirdly, the feature extraction and fusion module firstly extracts features of the encoded voxelized point cloud data to obtain a three-dimensional feature map, then carries out dimension conversion on the three-dimensional feature map to obtain a two-dimensional feature map, and finally carries out weighted fusion on all the two-dimensional feature maps to obtain a final feature map;
and step four, the target regression and detection module obtains a three-dimensional target detection frame according to the final feature map, wherein the three-dimensional target detection frame is a final three-dimensional target detection result.
In the second step, the feature code specifically includes: and taking the coordinate of each point in the voxelized point cloud data, the coordinate offset value of each point relative to the gravity center of the voxelized point cloud data, the coordinate offset value of each point relative to the positive center point of the voxelized point cloud data and the reflection intensity as the characteristic value of each point in the voxelized point cloud data.
The fourth step is specifically as follows: the target regression and detection module firstly convolves the final characteristic diagram to obtain a thermodynamic diagram, then extracts a thermodynamic diagram center point, regresses based on the thermodynamic diagram center point to obtain a center point, a length, a width, a height and a rotation angle of a three-dimensional target detection frame, and finally obtains the three-dimensional target detection frame, wherein the three-dimensional target detection frame is a final three-dimensional target detection result, and the thermodynamic diagram center point is a peak value which is larger than a set threshold value in the thermodynamic diagram.
The regression step of the central point of the three-dimensional target detection frame specifically comprises the following steps: and obtaining a Z value of the central point of the three-dimensional target detection frame through regression, and obtaining an X value and a Y value of the central point of the three-dimensional target detection frame through the combination of the X value and the Y value of the central point of the thermodynamic diagram and the predicted offset value through regression, so that the regression is completed.
In the first step, the voxel division specifically includes: the point cloud data was voxelized according to voxel scales of 0.1m x 0.4m, 0.2m x 0.4m, 0.4m x 0.4m, respectively.
In the third step, the feature extraction specifically comprises the following steps: firstly, extracting the characteristics of each coded voxelized point cloud data through the popular convolution and the sparse convolution in the convolutional neural network respectively to obtain a popular convolution result and a sparse convolution result, and then combining the popular convolution result and the sparse convolution result to obtain a three-dimensional characteristic diagram.
Compared with the prior art, the invention has the beneficial effects that:
1. according to the laser point cloud-based three-dimensional target detection method, the point cloud data of the three-dimensional target is subjected to voxel division by adopting different voxel scales to obtain a plurality of voxelized point cloud data, and then the voxelized point cloud data under the different voxel scales are subjected to feature coding, feature extraction and fusion in sequence, so that the structural features of the point cloud data can be guaranteed not to be lost, and the three-dimensional target detection precision is improved. Therefore, the invention improves the detection precision of the three-dimensional target.
2. The invention relates to a three-dimensional target detection method based on laser point cloud, which takes the coordinate of each point in voxelized point cloud data, the coordinate offset value of each point relative to the gravity center of the voxelized point cloud data, the coordinate offset value of each point relative to the positive center point of the voxelized point cloud data and the reflection intensity as the characteristic value of each point in the voxelized point cloud data. Therefore, the invention further improves the detection precision of the three-dimensional target.
3. According to the three-dimensional target detection method based on the laser point cloud, the final characteristic diagram is convolved to obtain the thermodynamic diagram, then the central point, the length, the width, the height and the rotation angle of the three-dimensional target detection frame are obtained by extracting the central point of the thermodynamic diagram and regressing based on the central point of the thermodynamic diagram, and finally the three-dimensional target detection frame is obtained to serve as a final three-dimensional target detection result. Therefore, the invention improves the three-dimensional target detection efficiency.
Drawings
FIG. 1 is a schematic structural diagram of the present invention.
FIG. 2 is a flow chart of the present invention.
In the figure, a voxel size division module 1, a feature coding module 2, a feature extraction and fusion module 3, a target regression and detection module 4, and a laser radar 5.
Detailed Description
The present invention will be further described with reference to the following embodiments.
Referring to fig. 1 and 2, the three-dimensional target detection system based on the laser point cloud comprises a voxel size division module 1, a feature coding module 2, a feature extraction and fusion module 3 and a target regression and detection module 4, wherein a signal output end of the voxel size division module 1 is connected with a signal input end of the target regression and detection module 4 sequentially through the feature coding module 2, the feature extraction and fusion module 3;
the voxel size division module 1 is used for obtaining point cloud data of a three-dimensional target, performing voxel division on the point cloud data by adopting different voxel scales to obtain a plurality of voxelized point cloud data, and then sending the voxelized point cloud data to the feature coding module 2;
the feature coding module 2 is used for firstly carrying out feature coding on the voxelized point cloud data and then sending the coded voxelized point cloud data to the feature extraction and fusion module 3;
the feature extraction and fusion module is used for sequentially performing feature extraction and weighted fusion on the encoded voxelized point cloud data to obtain a final feature map, and then sending the final feature map to the target regression and detection module 4;
and the target regression and detection module 4 is used for obtaining a three-dimensional target detection frame according to the final characteristic diagram.
The detection system further comprises a laser radar 5, and the signal input end of the voxel size dividing module 1 is connected with the signal output end of the laser radar 5.
A detection method of a three-dimensional target detection system based on laser point cloud sequentially comprises the following steps:
firstly, the voxel size division module 1 obtains point cloud data of a three-dimensional target from a laser radar 5, performs voxel division on the point cloud data by adopting different voxel scales to obtain a plurality of voxelized point cloud data, and then sends the voxelized point cloud data to the feature coding module 2;
secondly, the feature coding module 2 firstly carries out feature coding on the voxelized point cloud data and then sends the coded voxelized point cloud data to the feature extraction and fusion module 3;
thirdly, the feature extraction and fusion module 3 firstly extracts features of the encoded voxelized point cloud data to obtain a three-dimensional feature map, then carries out dimension conversion on the three-dimensional feature map to obtain a two-dimensional feature map, and finally carries out weighted fusion on all the two-dimensional feature maps to obtain a final feature map;
and step four, the target regression and detection module 4 obtains a three-dimensional target detection frame according to the final feature map, wherein the three-dimensional target detection frame is a final three-dimensional target detection result.
In the second step, the feature code specifically includes: and taking the coordinate of each point in the voxelized point cloud data, the coordinate offset value of each point relative to the gravity center of the voxelized point cloud data, the coordinate offset value of each point relative to the positive center point of the voxelized point cloud data and the reflection intensity as the characteristic value of each point in the voxelized point cloud data.
The fourth step is specifically as follows: the target regression and detection module 4 firstly convolves the final feature graph to obtain a thermodynamic diagram, then extracts a thermodynamic diagram center point, regresses based on the thermodynamic diagram center point to obtain a center point, a length, a width, a height and a rotation angle of a three-dimensional target detection frame, and finally obtains the three-dimensional target detection frame, wherein the three-dimensional target detection frame is a final three-dimensional target detection result, and the thermodynamic diagram center point is a peak value which is larger than a set threshold value on the thermodynamic diagram.
The regression step of the central point of the three-dimensional target detection frame specifically comprises the following steps: and obtaining a Z value of the central point of the three-dimensional target detection frame through regression, and obtaining an X value and a Y value of the central point of the three-dimensional target detection frame through the combination of the X value and the Y value of the central point of the thermodynamic diagram and the predicted offset value through regression, so that the regression is completed.
In the first step, the voxel division specifically includes: the point cloud data was voxelized according to voxel scales of 0.1m x 0.4m, 0.2m x 0.4m, 0.4m x 0.4m, respectively.
In the third step, the feature extraction specifically comprises the following steps: firstly, extracting the characteristics of each coded voxelized point cloud data through the popular convolution and the sparse convolution in the convolutional neural network respectively to obtain a popular convolution result and a sparse convolution result, and then combining the popular convolution result and the sparse convolution result to obtain a three-dimensional characteristic diagram.
The principle of the invention is illustrated as follows:
according to the laser point cloud-based three-dimensional target detection method, the gravity center of the voxelized point cloud data is the coordinate mean value of all points in the voxelized point cloud data, the coordinate offset value of each point relative to the gravity center of the voxelized point cloud data and the coordinate offset value of each point relative to the positive center point of the voxelized point cloud data can be used as the local aggregation characteristic of the points in the voxelized point cloud data and are encoded and convolved with the original characteristics of the points, namely the coordinates and the reflection intensity of each point, so that the extracted final characteristic graph can better learn the shape information of the three-dimensional target.
In the prior art, a three-dimensional target detection frame is obtained by generally adopting a mode of setting an anchor frame by a regional generation network RPN, in order to ensure the detection effect of the three-dimensional target detection frame, the number of the set anchor frames is far larger than the number of actual frames to be detected, and the excessive or repeated anchor frames need to be filtered by using a NMS non-maximum suppression algorithm, so that the detection speed is low, after a final characteristic diagram is obtained by the design, the final characteristic diagram is convolved to obtain a thermodynamic diagram, each point value on the thermodynamic diagram is in the range of 0-1, the higher the point value indicates that the point has higher probability of being the thermodynamic diagram central point, a thermodynamic diagram central point can be directly obtained by extracting a local peak value which is larger than a set threshold value on the thermodynamic diagram, then the central point, the length, the width, the height and the rotation angle of the three-dimensional target detection frame are obtained by regression based on the thermodynamic diagram central point, and, the detection efficiency is improved.
Example 1:
referring to fig. 1 and 2, the three-dimensional target detection system based on the laser point cloud comprises a voxel size division module 1, a feature coding module 2, a feature extraction and fusion module 3, a target regression and detection module 4 and a laser radar 5, wherein a signal output end of the laser radar 5 is connected with a signal input end of the target regression and detection module 4 sequentially through the voxel size division module 1, the feature coding module 2 and the feature extraction and fusion module 3;
the detection method of the three-dimensional target detection system based on the laser point cloud is sequentially carried out according to the following steps:
firstly, the voxel size division module 1 obtains point cloud data of a three-dimensional target from a laser radar 5, and voxelizes the point cloud data by respectively adopting a voxel scale of 0.1m x 0.4m, a voxel scale of 0.2m x 0.4m and a voxel scale of 0.4m x 0.4m to obtain 3 voxelized point cloud data, and then sends the 3 voxelized point cloud data to the feature coding module 2;
secondly, the feature coding module 2 firstly uses the coordinates of each point in the voxelized point cloud data, the coordinate offset value of each point relative to the gravity center of the voxelized point cloud data, the coordinate offset value of each point relative to the positive center point of the voxelized point cloud data and the reflection intensity as the feature value of each point in the voxelized point cloud data to perform feature coding on each voxelized point cloud data, and then sends 3 pieces of coded voxelized point cloud data to the feature extraction and fusion module 3;
thirdly, the feature extraction and fusion module 3 firstly extracts features of each coded voxelized point cloud data through popular convolution and sparse convolution in a convolutional neural network respectively to obtain popular convolutional layers and sparse convolutional layers, then combines the popular convolutional layers and the sparse convolutional layers to obtain a three-dimensional feature map, then conducts dimension conversion on the three-dimensional feature map to obtain a two-dimensional feature map, and finally conducts weighting fusion on the 3 two-dimensional feature maps to obtain a final feature map;
the target regression and detection module 4 firstly convolves the final feature graph to obtain a thermodynamic diagram, then extracts a peak value larger than a set threshold value on the thermodynamic diagram as a thermodynamic diagram central point, and regresses based on the thermodynamic diagram central point to obtain a central point, a length, a width, a height and a rotation angle of a three-dimensional target detection frame, and finally obtains the three-dimensional target detection frame which is a final three-dimensional target detection result;
the regression step of the central point of the three-dimensional target detection frame specifically comprises the following steps: and obtaining a Z value of the central point of the three-dimensional target detection frame through regression, and obtaining an X value and a Y value of the central point of the three-dimensional target detection frame through the combination of the X value and the Y value of the central point of the thermodynamic diagram and the predicted offset value through regression, so that the regression is completed.
Claims (8)
1. A three-dimensional target detection system based on laser point cloud is characterized in that:
the detection system comprises a voxel size division module (1), a feature coding module (2), a feature extraction and fusion module (3) and a target regression and detection module (4), wherein the signal output end of the voxel size division module (1) is connected with the signal input end of the target regression and detection module (4) through the feature coding module (2) and the feature extraction and fusion module (3) in sequence;
the voxel size division module (1) is used for obtaining point cloud data of a three-dimensional target and carrying out voxel division on the point cloud data by adopting different voxel scales to obtain a plurality of voxelized point cloud data, and then sending the voxelized point cloud data to the feature coding module (2);
the characteristic coding module (2) is used for firstly carrying out characteristic coding on the voxelized point cloud data and then sending the coded voxelized point cloud data to the characteristic extraction and fusion module (3);
the feature extraction and fusion module (3) is used for sequentially carrying out feature extraction and weighted fusion on the encoded voxelized point cloud data to obtain a final feature map, and then sending the final feature map to the target regression and detection module (4);
and the target regression and detection module (4) is used for obtaining a three-dimensional target detection frame according to the final characteristic diagram.
2. The laser point cloud-based three-dimensional target detection system of claim 1, wherein: the detection system further comprises a laser radar (5), and the signal input end of the voxel size dividing module (1) is connected with the signal output end of the laser radar (5).
3. The detection method of the laser point cloud-based three-dimensional target detection system according to claim 2, wherein:
the detection method sequentially comprises the following steps:
firstly, a voxel size division module (1) obtains point cloud data of a three-dimensional target from a laser radar (5) and performs voxel division on the point cloud data by adopting different voxel scales to obtain a plurality of voxelized point cloud data, and then sends the voxelized point cloud data to a feature coding module (2);
secondly, the characteristic coding module (2) firstly carries out characteristic coding on the voxelized point cloud data and then sends the coded voxelized point cloud data to the characteristic extraction and fusion module (3);
the feature extraction and fusion module (3) firstly extracts features of the encoded voxelized point cloud data to obtain a three-dimensional feature map, then carries out dimension conversion on the three-dimensional feature map to obtain a two-dimensional feature map, and finally carries out weighted fusion on all the two-dimensional feature maps to obtain a final feature map;
and step four, the target regression and detection module (4) obtains a three-dimensional target detection frame according to the final feature map, wherein the three-dimensional target detection frame is a final three-dimensional target detection result.
4. The detection method of the laser point cloud-based three-dimensional target detection system according to claim 3, wherein the detection method comprises the following steps: in the second step, the feature code specifically includes: and taking the coordinate of each point in the voxelized point cloud data, the coordinate offset value of each point relative to the gravity center of the voxelized point cloud data, the coordinate offset value of each point relative to the positive center point of the voxelized point cloud data and the reflection intensity as the characteristic value of each point in the voxelized point cloud data.
5. The detection method of the laser point cloud-based three-dimensional target detection system according to claim 3 or 4, wherein: the fourth step is specifically as follows: the target regression and detection module (4) firstly convolves the final feature graph to obtain a thermodynamic diagram, then extracts a thermodynamic diagram center point, regresses based on the thermodynamic diagram center point to obtain a center point, a length, a width, a height and a rotation angle of a three-dimensional target detection frame, and finally obtains the three-dimensional target detection frame, wherein the three-dimensional target detection frame is a final three-dimensional target detection result, and the thermodynamic diagram center point is a peak value which is larger than a set threshold value on the thermodynamic diagram.
6. The detection method of the laser point cloud-based three-dimensional target detection system according to claim 5, wherein: the regression step of the central point of the three-dimensional target detection frame specifically comprises the following steps: and obtaining a Z value of the central point of the three-dimensional target detection frame through regression, and obtaining an X value and a Y value of the central point of the three-dimensional target detection frame through the combination of the X value and the Y value of the central point of the thermodynamic diagram and the predicted offset value through regression, so that the regression is completed.
7. The detection method of the laser point cloud-based three-dimensional target detection system according to claim 3 or 4, wherein: in the first step, the voxel division specifically includes: the point cloud data was voxelized using voxel scales 0.1m x 0.4m, 0.2m x 0.4m, and 0.4m x 0.4m, respectively.
8. The detection method of the laser point cloud-based three-dimensional target detection system according to claim 3 or 4, wherein: in the third step, the feature extraction specifically comprises the following steps: firstly, extracting the characteristics of each coded voxelized point cloud data through the popular convolution and the sparse convolution in the convolutional neural network respectively to obtain a popular convolution result and a sparse convolution result, and then combining the popular convolution result and the sparse convolution result to obtain a three-dimensional characteristic diagram.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110004308.3A CN112731339A (en) | 2021-01-04 | 2021-01-04 | Three-dimensional target detection system based on laser point cloud and detection method thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110004308.3A CN112731339A (en) | 2021-01-04 | 2021-01-04 | Three-dimensional target detection system based on laser point cloud and detection method thereof |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112731339A true CN112731339A (en) | 2021-04-30 |
Family
ID=75589479
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110004308.3A Pending CN112731339A (en) | 2021-01-04 | 2021-01-04 | Three-dimensional target detection system based on laser point cloud and detection method thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112731339A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113378647A (en) * | 2021-05-18 | 2021-09-10 | 浙江工业大学 | Real-time rail obstacle detection method based on three-dimensional point cloud |
| CN113688738A (en) * | 2021-08-25 | 2021-11-23 | 北京交通大学 | Target identification system and method based on laser radar point cloud data |
| CN113706480A (en) * | 2021-08-13 | 2021-11-26 | 重庆邮电大学 | Point cloud 3D target detection method based on key point multi-scale feature fusion |
| CN113822244A (en) * | 2021-11-22 | 2021-12-21 | 之江实验室 | Point cloud three-dimensional target detection method and system for optimizing small target |
| CN113870318A (en) * | 2021-12-02 | 2021-12-31 | 之江实验室 | Moving target detection system and method based on multi-frame point cloud |
| CN114170126A (en) * | 2022-02-11 | 2022-03-11 | 智道网联科技(北京)有限公司 | Point cloud densification method and device and electronic equipment |
| CN114648579A (en) * | 2022-02-15 | 2022-06-21 | 浙江零跑科技股份有限公司 | Multi-branch input laser radar target detection method |
| CN115239548A (en) * | 2022-07-11 | 2022-10-25 | 小米汽车科技有限公司 | Target detection method, target detection device, electronic device, and medium |
| CN115588187A (en) * | 2022-12-13 | 2023-01-10 | 华南师范大学 | Pedestrian detection method, device and equipment based on three-dimensional point cloud and storage medium |
| CN115980702A (en) * | 2023-03-10 | 2023-04-18 | 安徽蔚来智驾科技有限公司 | Target false detection preventing method, device, driving device and medium |
| CN116486283A (en) * | 2023-01-09 | 2023-07-25 | 深圳优立全息科技有限公司 | Real-time point cloud target detection method and device based on voxel division |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110059608A (en) * | 2019-04-11 | 2019-07-26 | 腾讯科技（深圳）有限公司 | A kind of object detecting method, device, electronic equipment and storage medium |
| CN110222626A (en) * | 2019-06-03 | 2019-09-10 | 宁波智能装备研究院有限公司 | A kind of unmanned scene point cloud target mask method based on deep learning algorithm |
| CN111027581A (en) * | 2019-08-23 | 2020-04-17 | 中国地质大学（武汉） | 3D target detection method and system based on learnable codes |
| CN111199206A (en) * | 2019-12-30 | 2020-05-26 | 上海眼控科技股份有限公司 | Three-dimensional target detection method and device, computer equipment and storage medium |
| CN111862101A (en) * | 2020-07-15 | 2020-10-30 | 西安交通大学 | 3D point cloud semantic segmentation method under aerial view coding visual angle |
| CN112052860A (en) * | 2020-09-11 | 2020-12-08 | 中国人民解放军国防科技大学 | Three-dimensional target detection method and system |
-
2021
- 2021-01-04 CN CN202110004308.3A patent/CN112731339A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110059608A (en) * | 2019-04-11 | 2019-07-26 | 腾讯科技（深圳）有限公司 | A kind of object detecting method, device, electronic equipment and storage medium |
| CN110222626A (en) * | 2019-06-03 | 2019-09-10 | 宁波智能装备研究院有限公司 | A kind of unmanned scene point cloud target mask method based on deep learning algorithm |
| CN111027581A (en) * | 2019-08-23 | 2020-04-17 | 中国地质大学（武汉） | 3D target detection method and system based on learnable codes |
| CN111199206A (en) * | 2019-12-30 | 2020-05-26 | 上海眼控科技股份有限公司 | Three-dimensional target detection method and device, computer equipment and storage medium |
| CN111862101A (en) * | 2020-07-15 | 2020-10-30 | 西安交通大学 | 3D point cloud semantic segmentation method under aerial view coding visual angle |
| CN112052860A (en) * | 2020-09-11 | 2020-12-08 | 中国人民解放军国防科技大学 | Three-dimensional target detection method and system |
Non-Patent Citations (3)
| Title |
|---|
| H.KUANG等: ""Voxel-FPN :Multi-Scale Voxel Feature Aggregation for 3D Object Detection from LIDAR Point Clouds"", 《SENSORS》, vol. 20, no. 3, pages 1 - 18 * |
| XINGYI ZHOU 等: "Objects as Points", 《COMPUTER VISION AND PATTERN RECOGNITION》, pages 1 - 12 * |
| YIN ZHOU等: ""VoxelNet ：End-to-End Learning for PointCloud Based 3D Objcet Detection"", 《2018IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR).IEEE》, pages 1 - 10 * |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113378647A (en) * | 2021-05-18 | 2021-09-10 | 浙江工业大学 | Real-time rail obstacle detection method based on three-dimensional point cloud |
| CN113378647B (en) * | 2021-05-18 | 2024-03-29 | 浙江工业大学 | Real-time track obstacle detection method based on three-dimensional point cloud |
| CN113706480B (en) * | 2021-08-13 | 2022-12-09 | 重庆邮电大学 | Point cloud 3D target detection method based on key point multi-scale feature fusion |
| CN113706480A (en) * | 2021-08-13 | 2021-11-26 | 重庆邮电大学 | Point cloud 3D target detection method based on key point multi-scale feature fusion |
| CN113688738A (en) * | 2021-08-25 | 2021-11-23 | 北京交通大学 | Target identification system and method based on laser radar point cloud data |
| CN113688738B (en) * | 2021-08-25 | 2024-04-09 | 北京交通大学 | Target identification system and method based on laser radar point cloud data |
| CN113822244A (en) * | 2021-11-22 | 2021-12-21 | 之江实验室 | Point cloud three-dimensional target detection method and system for optimizing small target |
| CN113870318A (en) * | 2021-12-02 | 2021-12-31 | 之江实验室 | Moving target detection system and method based on multi-frame point cloud |
| WO2023098018A1 (en) * | 2021-12-02 | 2023-06-08 | 之江实验室 | Moving target detection system and method based on multi-frame point clouds |
| US11900618B2 (en) | 2021-12-02 | 2024-02-13 | Zhejiang Lab | System and method for detecting moving target based on multi-frame point cloud |
| CN114170126A (en) * | 2022-02-11 | 2022-03-11 | 智道网联科技(北京)有限公司 | Point cloud densification method and device and electronic equipment |
| CN114648579A (en) * | 2022-02-15 | 2022-06-21 | 浙江零跑科技股份有限公司 | Multi-branch input laser radar target detection method |
| CN115239548A (en) * | 2022-07-11 | 2022-10-25 | 小米汽车科技有限公司 | Target detection method, target detection device, electronic device, and medium |
| CN115588187A (en) * | 2022-12-13 | 2023-01-10 | 华南师范大学 | Pedestrian detection method, device and equipment based on three-dimensional point cloud and storage medium |
| CN116486283A (en) * | 2023-01-09 | 2023-07-25 | 深圳优立全息科技有限公司 | Real-time point cloud target detection method and device based on voxel division |
| CN115980702A (en) * | 2023-03-10 | 2023-04-18 | 安徽蔚来智驾科技有限公司 | Target false detection preventing method, device, driving device and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112731339A (en) | Three-dimensional target detection system based on laser point cloud and detection method thereof | |
| CN111832655B (en) | Multi-scale three-dimensional target detection method based on characteristic pyramid network | |
| CN108830280B (en) | Small target detection method based on regional nomination | |
| CN110111345B (en) | Attention network-based 3D point cloud segmentation method | |
| CN112241008B (en) | Method and system for object detection | |
| CN112288667B (en) | Three-dimensional target detection method based on fusion of laser radar and camera | |
| WO2022141718A1 (en) | Method and system for assisting point cloud-based object detection | |
| CN113536920B (en) | Semi-supervised three-dimensional point cloud target detection method | |
| CN112270332A (en) | Three-dimensional target detection method and system based on sub-stream sparse convolution | |
| CN113076804B (en) | Target detection method, device and system based on YOLOv4 improved algorithm | |
| CN103729846A (en) | LiDAR point cloud data edge detection method based on triangular irregular network | |
| CN116030445A (en) | Automatic driving real-time three-dimensional target detection method combining point cloud shape characteristics | |
| CN114299160A (en) | Dynamic detection method for pantograph running state based on fusion of vision and laser radar | |
| CN115984637A (en) | Time sequence fused point cloud 3D target detection method, system, terminal and medium | |
| CN116189147A (en) | YOLO-based three-dimensional point cloud low-power-consumption rapid target detection method | |
| CN116222577A (en) | Closed loop detection method, training method, system, electronic equipment and storage medium | |
| CN113989718A (en) | Human body target detection method facing radar signal heat map | |
| CN108345007B (en) | Obstacle identification method and device | |
| CN112801928A (en) | Attention mechanism-based millimeter wave radar and visual sensor fusion method | |
| CN108197613B (en) | Face detection optimization method based on deep convolution cascade network | |
| CN115731398A (en) | Point cloud and image fusion three-dimensional target detection method based on pseudo image | |
| WO2022017129A1 (en) | Target object detection method and apparatus, electronic device, and storage medium | |
| CN114821508A (en) | Road three-dimensional target detection method based on implicit context learning | |
| CN116664874B (en) | Single-stage fine-granularity light-weight point cloud 3D target detection system and method | |
| CN112883840B (en) | Power transmission line extraction method based on key point detection |
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 |