CN109948661B - 3D vehicle detection method based on multi-sensor fusion - Google Patents
3D vehicle detection method based on multi-sensor fusion Download PDFInfo
- Publication number
- CN109948661B CN109948661B CN201910144580.4A CN201910144580A CN109948661B CN 109948661 B CN109948661 B CN 109948661B CN 201910144580 A CN201910144580 A CN 201910144580A CN 109948661 B CN109948661 B CN 109948661B
- Authority
- CN
- China
- Prior art keywords
- point cloud
- vehicle
- frame
- laser radar
- anchor
- 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
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Image Analysis (AREA)
- Traffic Control Systems (AREA)
- Length Measuring Devices By Optical Means (AREA)
- Optical Radar Systems And Details Thereof (AREA)
Abstract
The invention discloses a 3D vehicle detection method based on multi-sensor fusion, which comprises the following steps: step 1, obtaining semantic information (namely RGB (red, green and blue) images) of a vehicle through a camera arranged on the vehicle, and scanning the surrounding environment of the vehicle through a laser radar positioned on the roof of the vehicle to obtain accurate depth information (namely laser radar point cloud) of the environment; step 2, preprocessing the laser radar point cloud, taking a Z axis [0,2.5] m according to the height of the automobile, and equally dividing the point cloud into 5 height slices along the Z axis direction; step 3, generating a 3D vehicle region of interest on the laser radar point cloud; step 4, respectively extracting the features of the processed radar point cloud and the processed RGB image and generating corresponding feature maps; step 5, respectively mapping the 3D vehicle interesting regions to a point cloud and a feature map of an RGB image; and 6, fusing the mapping part of the characteristic diagrams in the step 5, and finally realizing the 3D positioning and detection of the vehicle target.
Description
Technical Field
The invention belongs to the field of automatic driving, and particularly relates to a vehicle 3D detection method based on multi-sensor fusion.
Background
The intelligent vehicle is a complex system comprising technologies such as sensing, decision-making and control, environment sensing provides basic information for path planning and decision-making control, automobile detection is the most critical work in the environment sensing system of the autonomous vehicle, the mainstream obstacle detection sensors are a camera and a laser radar, the existing vehicle detection based on vision has a good effect, the cost of the camera is low, the texture and the color of a target can be obtained, and therefore the camera is used in the aspect of intelligent driving, but the camera is widely sensitive to illumination and shadow parts, cannot provide accurate and sufficient position information, and often causes the problems of low real-time performance or poor robustness and the like. The laser radar can obtain the target distance and the three-position information, the detection distance is far, the influence of illumination is avoided, the texture and the color of the target cannot be determined, and therefore the requirement for autonomous driving cannot be met by a single sensor. Therefore, the laser radar and the camera are adopted to perform data fusion to complete vehicle detection and tracking tasks, the dependence on the detection effect of a single sensor in vehicle detection is reduced, and a high 3D vehicle detection rate is obtained.
Disclosure of Invention
The invention aims to better detect surrounding vehicles so as to provide basic information for intelligent vehicle path planning and decision making, and provides a 3-dimensional (3D) vehicle detection method based on sensor fusion, which can obtain higher 3D vehicle detection rate.
The technical scheme adopted by the 3D vehicle detection method based on multi-sensor fusion provided by the invention comprises the following steps:
A3D vehicle detection method based on multi-sensor fusion comprises the following steps:
step 1, obtaining semantic information (namely RGB (red, green and blue) images) of a vehicle through a camera arranged on the vehicle, scanning the surrounding environment of the vehicle through a laser radar positioned on the roof of the vehicle, and obtaining accurate depth information (namely laser radar point cloud) of the environment;
step 2, preprocessing the laser radar point cloud, establishing a coordinate system which takes a contact point of the laser radar vertically downwards and the ground as an original point, the vehicle running direction as the positive direction of an X axis, the left side of a driver as the positive direction of a Y axis and the vertical ground upwards as the positive direction of a Z axis according to the height of the automobile, taking the Z axis [0,2.5] m, and equally dividing the point cloud into 5 height slices along the Z axis direction;
step 3, generating a 3D vehicle region of interest on the laser radar point cloud;
step 4, respectively extracting the features of the processed radar point cloud and the processed RGB image and generating corresponding feature maps;
step 5, respectively mapping the 3D vehicle interesting regions to a point cloud and a feature map of an RGB image;
and 6, fusing the mapping part of the characteristic diagrams in the step 5, and finally realizing the 3D positioning and detection of the vehicle target.
Further, the preprocessing of the step 2 comprises a processing method of a point cloud aerial view (BEV):
a Bird's Eye View (BEV) of the point cloud is obtained by projecting point cloud data onto a ground (Z = 0) 2D grid, and in order to obtain more detailed height information, a left and right position [ -40,40] m and a front position [0,70] m of the BEV are taken with the laser radar position as a center point. And according to the actual height of the automobile, taking a Z axis [0,2.5] m, equally dividing the point cloud into 5 height slices along the Z axis direction, projecting each slice to a ground (Z = 0) 2D grid, and taking the height characteristic corresponding to each slice as the maximum height value of the point cloud data projected to the grid map. The point cloud density M refers to the number of point clouds of each cell, and the value of each grid is normalized:
and N is the number of point clouds in the unit grid map.
Further, the specific steps of step 3 are:
generating a 3D vehicle region of interest on a point cloud for classifying and positioning targets, taking a Bird's Eye View (BEV) as an input, generating a series of 3D candidate frames before, removing empty frames to reduce the calculation amount, assigning a binary label to the content of each remaining frame, namely a positive label represents a target vehicle, a negative label represents a background, and assigning a positive label to two types of anchor frames by calculating the IOU overlapping size between the anchor frame and a real boundary frame:
1) The anchor box with the highest IOU overlap with a real bounding box (less than 0.5),
2) An anchor box that overlaps any IOU with a true bounding box greater than 0.5.
And one real bounding box may assign positive labels to multiple anchor boxes. Assigning negative labels (backgrounds) to anchor boxes whose IOU of all real bounding boxes is lower than 0.3, and non-positive and non-negative anchor boxes have no effect on training targets, so we ignore in subsequent processing. After the above-mentioned positively labeled anchor frame is obtained, it is subjected to preliminary 3D regression optimization, assuming that each 3D prediction frame is represented by (x, y, z, h, w, D), (x, y, z) represents the center point of the frame, and (h, w, D) represents the size of the frame. The centroid and the size of the 3D frame in the laser radar coordinate system are distinguished and preliminarily positioned for the ROI generated by mapping on the feature map later by calculating the difference (delta x, delta y, delta z, delta h, delta w and delta D) between the boundary frame of the target area with the foreground and the real boundary frame in the central point and the size. For 3D anchor frame (x) a ,y a ,z a ,h a ,w a ,d a ) Representation, 3D true bounding Box by (x) * ,y * ,z * ,h * ,w * ,d * ) Denotes, t i Denotes the offset of the prediction frame with respect to the 3D anchor frame, let t i 6 parameterized coordinates as t i =(t x ,t y ,t z ,t h ,t w ,t d ),Represents the offset of the 3D real bounding box relative to the 3D anchor frame, and is ^ based>A parameterized coordinate ofThen there are:
t x =(x-x a )/h a t y =(y-ya)/w a
t z =(z-z a )/d a t h =log(h/h a )
t w =log(w/w a )t d =log(d/d a )
the cross-entropy function was used to calculate the target object loss:where n is the number of bounding boxes present in the target region.
And performing 3D frame regression by calculating the difference between the centroid and the size between the 3D anchor frame and the 3D real boundary frame, and finally outputting a 3D region of interest in the point cloud.
Further, the specific process of step 4 includes:
step 4.1, assuming the input RGB image or BEV image size is H × W × D, the first three convolutional layers of the VGG-16 network are used in the downsampling stage, resulting in an output signature with a resolution 8 times smaller than its corresponding input, at which stage the output size of the signature is H × W × D
And 4.2, performing 2X up-sampling on the feature map of high-level low-resolution semantic information (including laser radar point cloud and RGB (red, green and blue) images), ensuring that the feature map of a down-sampling stage corresponding to the up-sampling has the same size, and performing convolution fusion on the feature map by 3X3, thereby obtaining a full-resolution feature map in the last layer of the feature extraction framework.
Further, the specific method of steps 5 and 6 is as follows:
and (4) obtaining a 3D region of interest on a bird-eye view (BEV) of the laser radar point cloud obtained in the step (3), respectively mapping the region of interest obtained on the radar point cloud onto the feature maps of the radar point cloud and the RGB image according to the coordinate relation between the laser radar point cloud and the RGB image, and finally obtaining the coordinate position of a corresponding frame on the feature map, wherein the size of the frame finally mapped on the feature map is different, so that the fusion processing cannot be carried out, the size of the obtained feature map is fixed to be 3X3, and then the pixel average fusion is carried out on the feature maps obtained by mapping in the BEV and the RGB.
1. The invention has the beneficial effects that: the invention uses the laser radar and the camera to sense the surrounding environment, can fully utilize the data acquired by the laser radar, has accurate depth information and the camera can reserve the advantage of more detailed semantic information. The accuracy of the 3D detection of the surrounding vehicle is greatly improved.
2. The traditional method for detecting the vehicle by using the single sensor has limited acquired information and is influenced by the performance of the single sensor.
3. According to the invention, the vehicle interested target is extracted firstly, and then the pixel average fusion processing is carried out on the part of characteristic images, so that the calculated amount is greatly reduced, and the real-time property of vehicle detection is improved.
Drawings
FIG. 1 is a flow chart of a method for vehicle detection based on multi-sensor fusion in accordance with the present invention;
FIG. 2 is a Bird's Eye View (BEV) of a point cloud equally divided into 5 height slices along the Z-axis direction;
(a) shows a radar point cloud along the Z-axis [0,0.5], (b) shows a radar point cloud along the Z-axis [0.5,1.0], (c) shows a radar point cloud along the Z-axis [1.0,1.5], (d) shows a radar point cloud along the Z-axis [1.5,2.0], (e) shows a radar point cloud along the Z-axis [2.0,2.5 ];
fig. 3 is a frame for extracting features of a point cloud and RGB image.
Detailed Description
The invention will be further explained with reference to the drawings.
The invention provides a 3D vehicle detection method based on multi-sensor fusion, wherein the detection flow chart is shown in figure 1 and specifically comprises the following steps:
(1) The method comprises the steps of collecting point cloud data through a laser radar, collecting RGB image information through a camera, preprocessing the collected point cloud, inputting a Bird's Eye View (BEV) of the point cloud as point cloud data, projecting the point cloud data to a ground (Z = 0) 2D grid to obtain more detailed height information, and taking the position of the laser radar as a central point, wherein the position is the left and right positions [ -40,40] m of the BEV, and the position is the front position [0,70] m. And according to the actual height of the automobile, taking a Z axis [0,2.5] m, equally dividing the point cloud into 5 height slices along the Z axis direction, projecting each slice to a ground (Z = 0) 2D grid, and taking the height characteristic corresponding to each slice as the maximum height value of the point cloud data projected to the grid map. The point cloud density M refers to the number of point clouds of each cell, and the value of each grid is normalized:
and N is the number of point clouds in the unit grid map.
(2) Generating a 3D vehicle region of interest on the point cloud for classification and positioning of targets, taking a Bird's Eye View (BEV) as an input, generating a series of 3D candidate frames before, removing empty frames in order to reduce the calculation amount, assigning a binary label to the content of each frame left, which is the target vehicle or background, and assigning a positive label to two types of anchor frames by calculating the IOU overlap size between the anchor frame and a real boundary frame:
(1) anchor boxes (less than 0.5) with highest IOU overlap with a real bounding box;
(2) an anchor box that overlaps any IOU with a true bounding box greater than 0.5.
And one real bounding box may assign positive labels to multiple anchor boxes. Assigning negative labels (backgrounds) to anchor boxes whose IOU of all real bounding boxes is below 0.3, and non-positive and non-negative anchor boxes have no effect on the training target and are ignored in subsequent processing. After the positive labeled anchor box is obtained, it is subjected to preliminary 3D regression optimization, assuming that each 3D prediction box is represented by (x, y, z, h, w, D) tableThe (x, y, z) represents the center point of the box, and (h, w, d) represents the size of the box. The center of mass and the size of the 3D frame in the laser radar coordinate system are distinguished and preliminarily positioned for the ROI generated by mapping on the feature map later by calculating the difference (delta x, delta y, delta z, delta h, delta w, delta D) between the foreground ROI and the real mania in the central point and the size. For 3D anchor frame (x) a ,y a ,z a ,h a ,w a ,d a ) Representation, 3D true bounding Box by (x) * ,y * ,z * ,h * ,w * ,d * ) Denotes, t i Representing the offset of the prediction frame relative to the 3D anchor frame, then its 6 parameterized coordinates are t i =(t x ,t y ,t z ,t h ,t w ,t d ),Representing the offset of the 3D real bounding box relative to the 3D anchor box, then its 6 parameterized coordinates are ^ 4>Then there are:
t x =(x-x a )/h a t y =(y-y a )/w a
t z =(z-z a )/d a t h =log(h/h a )
t w =log(w/w a ) t d =log(d/d a )
and performing 3D frame regression by calculating the difference between the centroid and the size between the 3D anchor frame and the 3D real boundary frame, and finally outputting a 3D region of interest in the point cloud.
(3) A bird's-eye view (BEV) of the point cloud is input as point cloud data, which is obtained by projecting the point cloud data onto a ground (Z = 0) 2D grid, and in order to obtain more detailed height information, a left-right position [ -40,40] m and a front position [0,70] m of the BEV are taken with the laser radar position as a center point. And according to the actual height of the automobile, taking a Z axis [0,2.5] m, equally dividing the point cloud into 5 height slices along the Z axis direction, projecting each slice to a ground (Z = 0) 2D grid, and taking the height characteristic corresponding to each slice as the maximum height value of the point cloud data projected to the grid map. The point cloud density M refers to the number of point clouds of each cell, and the value of each grid is normalized:
and N is the number of point clouds in the unit grid map.
(4) As shown in fig. 3, in order to fully utilize the information of the original lowest-layer feature map, the upper-layer feature upsampling and the bottom-layer information are fused by a 3X3 convolution operation. To obtain rich feature information and high resolution images. The feature extractor is based on the VGG-16 architecture. Assuming that the input RGB image or BEV map is H × W × D in size, using the first three convolutional layers of the VGG-16 network in the downsampling stage results in an output feature map with a resolution 8 times smaller than its corresponding input, so that at this stage the output size of the feature map isThe downsampled feature map is convolved by a convolution kernel 1X1 with some channels with corresponding upsampled phase feature maps, so that a 3X3 convolution fusion can be performed, resulting in a full resolution feature map in the last layer of the feature extraction framework.
(5) And (3) respectively mapping the 3D region of interest in the point cloud (2) to the feature maps of the point cloud and the RGB image, and obtaining the coordinate position of the corresponding frame on the feature map according to the coordinate conversion relation between the BEV and the RGB image. However, since the size of the frame obtained by mapping the feature map finally is different, the fusion processing cannot be performed, so that the size of the obtained feature map is fixed to be 3X3, and then the feature maps obtained by mapping in BEV and RGB are fused. And finally determining the position and the size of the surrounding vehicle.
The above-listed detailed description is only a specific description of a possible embodiment of the present invention, and they are not intended to limit the scope of the present invention, and equivalent embodiments or modifications made without departing from the technical spirit of the present invention should be included in the scope of the present invention.
Claims (7)
1. A3D vehicle detection method based on multi-sensor fusion is characterized by comprising the following steps:
step 1, obtaining an RGB image of a vehicle and laser radar point cloud information of the surrounding environment of the vehicle;
step 2, preprocessing the laser radar point cloud information, taking a Z axis [0,2.5] m according to the height of the automobile, and equally dividing the laser radar point cloud into 5 height slices along the Z axis direction;
step 3, generating a 3D vehicle region of interest on the laser radar point cloud; the method comprises the following specific steps:
taking a point cloud aerial view as an input, generating a series of 3D candidate frames before, removing the empty candidate frames, allocating a binary label to the content of each remaining candidate frame, namely a positive label represents a target vehicle and a negative label represents a background, and allocating a positive label to the following two types of anchor frames by calculating the IOU overlapping size between the anchor frame and a real boundary frame:
1) An anchor frame having an IOU overlap of up to less than 0.5 with a real bounding box;
2) An anchor box overlapping any real bounding box with an IOU greater than 0.5;
assigning negative labels to anchor frames with IOUs of all real bounding boxes lower than 0.3, wherein the anchor frames which are not positive or negative have no effect on the training target and are ignored in subsequent processing;
after obtaining the anchor frame with the positive label, performing preliminary 3D regression optimization on the anchor frame, and assuming that each 3D prediction frame is represented by (x, y, z, h, w, D), (x, y, z) represents the center point of the frame, and (h, w, D) represents the size of the frame; in a laser radar coordinate system, distinguishing and primarily positioning the ROI generated by mapping on a feature map later by calculating the difference between the foreground ROI and a real boundary frame in the central point and the size, namely (delta x, delta y, delta z, delta h, delta w and delta d); for 3D anchor frame (x) a ,y a ,z a ,h a ,w a ,d a ) Representation, 3D true bounding Box by (x) * ,y * ,z * ,h * ,w * ,d * ) Denotes, t i Represents the offset of the prediction frame relative to the 3D anchor frame, and has 6 parameterized coordinates t i =(t x ,t y ,t z ,t h ,t w ,t d ),Represents the offset of the 3D real bounding box relative to the 3D anchor frame, and sets its 6 parameterized coordinates to ^ 6>Then there are:
t x =(x-x a )/h a t y =(y-y a )/w a
t z =(z-z a )/d a t h =log(h/h a )
t w =log(w/w a ) t d =log(d/d a )
wherein n is the number of bounding boxes in the target area;
performing 3D frame regression by calculating the difference between the centroid and the size between the 3D anchor frame and the 3D real bounding box, and finally outputting a 3D region of interest in the point cloud;
step 4, respectively extracting the features of the processed radar point cloud and the processed RGB image and generating corresponding feature maps;
step 5, respectively mapping the 3D vehicle region of interest to a radar point cloud and a feature map of an RGB image;
and 6, fusing the mapping part of the characteristic graphs in the step 5, and finally realizing the 3D positioning and detection of the vehicle target.
2. The multi-sensor fusion-based 3D vehicle detection method according to claim 1, wherein in step 1, the RGB images are obtained by a camera mounted on a vehicle; and the laser radar point cloud scans and acquires the surrounding environment through a laser radar positioned on the roof of the vehicle.
3. The multi-sensor fusion-based 3D vehicle detection method according to claim 1, wherein the preprocessing method in step 2 comprises a processing method of a point cloud aerial view, wherein the point cloud aerial view is obtained by projecting point cloud data to a ground (Z = 0) 2D grid.
4. The multi-sensor fusion-based 3D vehicle detection method according to claim 3, wherein the processing method of the point cloud aerial view is as follows:
taking the position of a laser radar as a central point, taking the left and right positions [ -40,40] m and the front position [0,70] m of a point cloud aerial view, taking the Z axis [0,2.5] m according to the actual height of an automobile, equally dividing the point cloud into 5 height slices along the Z axis direction, projecting each slice to a ground (Z = 0) 2D grid, and taking the height characteristic corresponding to each slice as the maximum height value of the point cloud data projected to the grid map; the point cloud density M refers to the number of point clouds of each unit grid, and the value of each grid is normalized:
and N is the number of point clouds in the unit grid map.
5. The multi-sensor fusion based 3D vehicle detection method according to claim 1, wherein the specific steps of step 4 comprise:
step 4.1, assuming that the size of the input RGB image or point cloud aerial view is H × W × D, using the first three convolution layers of the VGG-16 network in the down-sampling stage results in the resolution of the output feature map being 8 times smaller than its corresponding input, at this stage, the output size of the feature map is set to be H × W × D
Step 4.2, performing 2X up-sampling on the feature map of the semantic information with high-level low resolution, ensuring that the feature map size of a down-sampling stage corresponding to the up-sampling is the same, performing convolution fusion on the feature map 3X3, and obtaining a full-resolution feature map in the last layer of the feature extraction frame; the semantic information comprises laser radar point cloud and RGB images.
6. The multi-sensor fusion-based 3D vehicle detection method according to claim 1, wherein the specific mapping method of step 5 is as follows: and respectively mapping the interested 3D areas acquired on the radar point cloud to the feature maps of the radar point cloud and the RGB image according to the corresponding coordinate relation between the radar point cloud and the RGB image.
7. The multi-sensor fusion-based 3D vehicle detection method according to claim 6, wherein the specific fusion method of step 6 comprises the following steps: and (5) carrying out pixel average fusion on the radar point cloud obtained in the step (5) and the feature image obtained by mapping the RGB image.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910144580.4A CN109948661B (en) | 2019-02-27 | 2019-02-27 | 3D vehicle detection method based on multi-sensor fusion |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910144580.4A CN109948661B (en) | 2019-02-27 | 2019-02-27 | 3D vehicle detection method based on multi-sensor fusion |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109948661A CN109948661A (en) | 2019-06-28 |
CN109948661B true CN109948661B (en) | 2023-04-07 |
Family
ID=67006948
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910144580.4A Active CN109948661B (en) | 2019-02-27 | 2019-02-27 | 3D vehicle detection method based on multi-sensor fusion |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109948661B (en) |
Families Citing this family (41)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11165462B2 (en) * | 2018-11-07 | 2021-11-02 | Samsung Electronics Co., Ltd. | Motion assisted leakage removal for radar applications |
US10699563B1 (en) * | 2019-07-12 | 2020-06-30 | GM Global Technology Operations LLC | Multi-sensor multi-object tracking |
CN110363158B (en) * | 2019-07-17 | 2021-05-25 | 浙江大学 | Millimeter wave radar and visual cooperative target detection and identification method based on neural network |
CN110458112B (en) * | 2019-08-14 | 2020-11-20 | 上海眼控科技股份有限公司 | Vehicle detection method and device, computer equipment and readable storage medium |
CN110543858A (en) * | 2019-09-05 | 2019-12-06 | 西北工业大学 | Multi-mode self-adaptive fusion three-dimensional target detection method |
CN110738121A (en) * | 2019-09-17 | 2020-01-31 | 北京科技大学 | front vehicle detection method and detection system |
CN110827202A (en) * | 2019-11-07 | 2020-02-21 | 上海眼控科技股份有限公司 | Target detection method, target detection device, computer equipment and storage medium |
CN110991534A (en) * | 2019-12-03 | 2020-04-10 | 上海眼控科技股份有限公司 | Point cloud data processing method, device, equipment and computer readable storage medium |
CN110929692B (en) * | 2019-12-11 | 2022-05-24 | 中国科学院长春光学精密机械与物理研究所 | Three-dimensional target detection method and device based on multi-sensor information fusion |
CN111144304A (en) * | 2019-12-26 | 2020-05-12 | 上海眼控科技股份有限公司 | Vehicle target detection model generation method, vehicle target detection method and device |
CN113678136A (en) * | 2019-12-30 | 2021-11-19 | 深圳元戎启行科技有限公司 | Obstacle detection method and device based on unmanned technology and computer equipment |
CN111209825B (en) * | 2019-12-31 | 2022-07-01 | 武汉中海庭数据技术有限公司 | Method and device for dynamic target 3D detection |
US11227404B2 (en) * | 2020-02-25 | 2022-01-18 | Zebra Technologies Corporation | Transporter segmentation for data capture system |
CN111291714A (en) * | 2020-02-27 | 2020-06-16 | 同济大学 | Vehicle detection method based on monocular vision and laser radar fusion |
CN113378605B (en) * | 2020-03-10 | 2024-04-09 | 北京京东乾石科技有限公司 | Multi-source information fusion method and device, electronic equipment and storage medium |
CN111239706B (en) * | 2020-03-30 | 2021-10-01 | 许昌泛网信通科技有限公司 | Laser radar data processing method |
CN111476822B (en) * | 2020-04-08 | 2023-04-18 | 浙江大学 | Laser radar target detection and motion tracking method based on scene flow |
CN111583663B (en) * | 2020-04-26 | 2022-07-12 | 宁波吉利汽车研究开发有限公司 | Monocular perception correction method and device based on sparse point cloud and storage medium |
CN111352112B (en) * | 2020-05-08 | 2022-11-29 | 泉州装备制造研究所 | Target detection method based on vision, laser radar and millimeter wave radar |
CN113674346B (en) * | 2020-05-14 | 2024-04-16 | 北京京东乾石科技有限公司 | Image detection method, image detection device, electronic equipment and computer readable storage medium |
CN113705279B (en) * | 2020-05-21 | 2022-07-08 | 阿波罗智联(北京)科技有限公司 | Method and device for identifying position of target object |
CN113763465A (en) * | 2020-06-02 | 2021-12-07 | 中移(成都)信息通信科技有限公司 | Garbage determination system, model training method, determination method and determination device |
CN111833358A (en) * | 2020-06-26 | 2020-10-27 | 中国人民解放军32802部队 | Semantic segmentation method and system based on 3D-YOLO |
CN112001226B (en) * | 2020-07-07 | 2024-05-28 | 中科曙光(南京)计算技术有限公司 | Unmanned 3D target detection method, device and storage medium |
CN113761999B (en) * | 2020-09-07 | 2024-03-05 | 北京京东乾石科技有限公司 | Target detection method and device, electronic equipment and storage medium |
CN112183393A (en) * | 2020-09-30 | 2021-01-05 | 深兰人工智能(深圳)有限公司 | Laser radar point cloud target detection method, system and device |
CN112287859A (en) * | 2020-11-03 | 2021-01-29 | 北京京东乾石科技有限公司 | Object recognition method, device and system, computer readable storage medium |
CN112184539A (en) * | 2020-11-27 | 2021-01-05 | 深兰人工智能(深圳)有限公司 | Point cloud data processing method and device |
CN114556427A (en) * | 2020-12-16 | 2022-05-27 | 深圳市大疆创新科技有限公司 | Point cloud processing method, point cloud processing apparatus, movable platform, and computer storage medium |
CN112711034B (en) * | 2020-12-22 | 2022-10-14 | 中国第一汽车股份有限公司 | Object detection method, device and equipment |
US11482007B2 (en) | 2021-02-10 | 2022-10-25 | Ford Global Technologies, Llc | Event-based vehicle pose estimation using monochromatic imaging |
CN112946689A (en) * | 2021-03-08 | 2021-06-11 | 苏州岭纬智能科技有限公司 | Integrated laser radar system and detection method thereof |
CN112990229A (en) * | 2021-03-11 | 2021-06-18 | 上海交通大学 | Multi-modal 3D target detection method, system, terminal and medium |
CN113011317B (en) * | 2021-03-16 | 2022-06-14 | 青岛科技大学 | Three-dimensional target detection method and detection device |
CN113128348B (en) * | 2021-03-25 | 2023-11-24 | 西安电子科技大学 | Laser radar target detection method and system integrating semantic information |
CN113222111A (en) * | 2021-04-01 | 2021-08-06 | 上海智能网联汽车技术中心有限公司 | Automatic driving 4D perception method, system and medium suitable for all-weather environment |
CN113192091B (en) * | 2021-05-11 | 2021-10-22 | 紫清智行科技(北京)有限公司 | Long-distance target sensing method based on laser radar and camera fusion |
CN113408454B (en) * | 2021-06-29 | 2024-02-06 | 上海高德威智能交通系统有限公司 | Traffic target detection method, device, electronic equipment and detection system |
CN113655497B (en) * | 2021-08-30 | 2023-10-27 | 杭州视光半导体科技有限公司 | Method for scanning region of interest based on FMCW solid-state scanning laser radar |
CN114463660A (en) * | 2021-12-14 | 2022-05-10 | 江苏航天大为科技股份有限公司 | Vehicle type judging method based on video radar fusion perception |
CN115909034A (en) * | 2022-11-29 | 2023-04-04 | 白城师范学院 | Point cloud target identification method and device based on scene density perception and storage medium |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107533630A (en) * | 2015-01-20 | 2018-01-02 | 索菲斯研究股份有限公司 | For the real time machine vision of remote sense and wagon control and put cloud analysis |
CN109100741A (en) * | 2018-06-11 | 2018-12-28 | 长安大学 | A kind of object detection method based on 3D laser radar and image data |
-
2019
- 2019-02-27 CN CN201910144580.4A patent/CN109948661B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107533630A (en) * | 2015-01-20 | 2018-01-02 | 索菲斯研究股份有限公司 | For the real time machine vision of remote sense and wagon control and put cloud analysis |
CN109100741A (en) * | 2018-06-11 | 2018-12-28 | 长安大学 | A kind of object detection method based on 3D laser radar and image data |
Also Published As
Publication number | Publication date |
---|---|
CN109948661A (en) | 2019-06-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109948661B (en) | 3D vehicle detection method based on multi-sensor fusion | |
CN112287860B (en) | Training method and device of object recognition model, and object recognition method and system | |
CN111583337A (en) | Omnibearing obstacle detection method based on multi-sensor fusion | |
CN111209825B (en) | Method and device for dynamic target 3D detection | |
CN113192091B (en) | Long-distance target sensing method based on laser radar and camera fusion | |
CN112825192B (en) | Object identification system and method based on machine learning | |
CN115049700A (en) | Target detection method and device | |
Wang et al. | An overview of 3d object detection | |
CN112097732A (en) | Binocular camera-based three-dimensional distance measurement method, system, equipment and readable storage medium | |
CN113658257B (en) | Unmanned equipment positioning method, device, equipment and storage medium | |
CN115187964A (en) | Automatic driving decision-making method based on multi-sensor data fusion and SoC chip | |
CN114331986A (en) | Dam crack identification and measurement method based on unmanned aerial vehicle vision | |
TWI745204B (en) | High-efficiency LiDAR object detection method based on deep learning | |
CN112287859A (en) | Object recognition method, device and system, computer readable storage medium | |
CN116486287A (en) | Target detection method and system based on environment self-adaptive robot vision system | |
CN117274749B (en) | Fused 3D target detection method based on 4D millimeter wave radar and image | |
CN113408324A (en) | Target detection method, device and system and advanced driving assistance system | |
CN113095152A (en) | Lane line detection method and system based on regression | |
CN112395962A (en) | Data augmentation method and device, and object identification method and system | |
CN117111055A (en) | Vehicle state sensing method based on thunder fusion | |
CN115457358A (en) | Image and point cloud fusion processing method and device and unmanned vehicle | |
TWI673190B (en) | Vehicle detection method based on optical radar | |
CN113219472B (en) | Ranging system and method | |
Eraqi et al. | Static free space detection with laser scanner using occupancy grid maps | |
CN113611008B (en) | Vehicle driving scene acquisition method, device, equipment and medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |