CN109543601A - A kind of unmanned vehicle object detection method based on multi-modal deep learning - Google Patents
A kind of unmanned vehicle object detection method based on multi-modal deep learning Download PDFInfo
- Publication number
- CN109543601A CN109543601A CN201811388553.3A CN201811388553A CN109543601A CN 109543601 A CN109543601 A CN 109543601A CN 201811388553 A CN201811388553 A CN 201811388553A CN 109543601 A CN109543601 A CN 109543601A
- Authority
- CN
- China
- Prior art keywords
- laser radar
- image
- unmanned vehicle
- deep learning
- object detection
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
- G06V20/584—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of vehicle lights or traffic lights
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10028—Range image; Depth image; 3D point clouds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
- G06T2207/10044—Radar image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Traffic Control Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of unmanned vehicle object detection methods based on multi-modal deep learning, comprising the following steps: (1) data acquire;(2) data are synchronous;(3) laser radar point cloud data is handled;(4) feature extraction;(5) tagsort returns;(6) it projects;(7) redundancy candidate frame is removed;(8) it corrects;(9) three dimensional detection frame is exported.The advantage of present invention combination multisensor overcomes the shortcomings of that image modalities to lane scene ability to express, cope with the scene perception under the complex scene that may be faced in unmanned vehicle traveling.
Description
Technical field
The present invention relates to the technical fields of pilotless target detection, are based on multi-modal depth more particularly to one kind
The unmanned vehicle object detection method of habit.
Background technique
Since automobile is born, the characteristics of convenience and rapidity, is that everybody is obvious to all, but along with vapour
Vehicle safety problem is also always the emphasis of car manufactures and consumer's concern.Avoiding road barricade in time is road vehicle peace
Full important prerequisite, so how to accomplish to allow vehicle can " identification " present road situation either dynamic object or static object
Body just becomes one ring of key of intelligent vehicle scene understanding.
The prior art is usually to use image modalities, but only be difficult to do intelligent vehicle road scene using image modalities
One accurately reproduces and analyzes out, and camera has larger parameter variation when illumination variation is big, to make effect by shadow
It rings, and ability to express is poor under the scenes weather such as rainy day, snowy day.It therefore, can be by increasing millimetre-wave radar, laser thunder
Up to the modal information of equal multisensors, the disadvantage of camera information deficiency is made up, so that intelligent vehicle can prejudge under various circumstances
The position of obstacle.
The patent of Publication No. CN107169421A describes a kind of car steering field based on depth convolutional neural networks
Scape object detection method, but the patent there is no solve camera for light sensitive, bad weather when effect it is very poor lack
Point can not provide when driving reliable scene perception information in unmanned vehicle, be only the framework for changing CNN network, improve
Detection effect when single image mode is as input.
The patent of Publication No. CN102253391A describes a kind of pedestrian target track side based on multilasered optical radar
Method detects pedestrian using the sensor network of multiple line laser radars composition, but the patent is provided only for pedestrian's
Detection, is unable to satisfy unmanned vehicle when driving for the demand of scene perception, and since the point cloud of laser radar is very sparse, far
Not as good as image data, so that it is extremely difficult in space search target.
Summary of the invention
In order to solve the above technical problems, the present invention provides a kind of advantage of combination multisensor, overcome image modalities to vehicle
The deficiency of road scene ability to express copes with the base of the scene perception under the complex scene that may be faced in unmanned vehicle traveling
In the unmanned vehicle object detection method of multi-modal deep learning.
A kind of unmanned vehicle object detection method based on multi-modal deep learning, comprising the following steps:
(1) data acquire: utilizing the image data and laser thunder of in-vehicle camera and mobile lidar acquisition monocular camera
The point cloud data reached;
(2) it handles laser radar point cloud data: looking down visual angle by look down that visual angle projection algorithm obtains laser radar
Figure;
(3) feature extraction: the deep layer for extracting laser radar top view and image data respectively using convolutional neural networks is special
Sign;
(4) tagsort returns, and respectively obtains the candidate frame of laser radar and the multiple targets of image data;
(5) it projects: laser radar being looked down into visual angle figure and the respective candidate frame of image data projects to the feature of same specification
Figure;
(6) it removes redundancy candidate frame: using laser radar candidate frame as standard, according to candidate frame central threshold range, abandoning
Redundancy anchor in image data does not occur in laser radar such as but the anchor of candidate frame occurs in image data, then deletes;
(7) the nomination correction of image aspects avoids differing biggish laser radar target nomination with practical;
(8) three dimensional detection frame is exported: the laser radar point cloud data array and reservation of rule-based input convolutional network
Depth information is returned out together by loss function to be detected the coordinate of vehicle and saves.
Wherein, the loss function be intersect entropy function and image section loss function multinomial and:
Wherein, Nl、NcRespectively represent the number of training of laser radar and image;ynFor the mark value of n-th of sample, it is
Vehicle is represented when 1, represents other objects when being 0,It is convolutional neural networks for the predicted value of sample;W is net
Parameter in network, multinomial first half are exactly the loss function definition about laser radar, and aft section is damaged about image
Considering for function is lost, image impairment function loses L by confidence levelconfL is lost with positioninglocIt forms, x represents prediction block in formula
Whether matched with true frame, c represents the matching degree of prediction block Yu true frame, and α is balance factor, when cross validation
Value is that 1, l represents prediction block, and g represents true frame, also represent the finally target detection classification level of confidence that returns out and
Whether the accurate positioning of each coordinate points of target detection, multiple risks function constraint is exactly to fit close to label data
True value improves the performance level of network entirety whereby.
The invention also includes data synchronizing steps, by the image of in-vehicle camera and mobile lidar acquisition monocular camera
The point cloud data of data and laser radar synchronizes processing.
The coordinate quantity of the detected vehicle is 8.
The tagsort returns, the sub-step including following non-sequential execution:
A. the feature extracted from laser radar point cloud data is sent into RPN network, mentioned feature is classified and returned
Return, obtain the frame coordinate value of multiple targets under 3D coordinate, by converting the candidate frame for obtaining 2D and looking down under visual angle;
B. the extracted feature of institute it will be sent into RPN network from image data, mentioned feature is classified and returned,
Obtain the candidate frame of multiple targets on the plane of delineation.
Described looks down visual angle projection algorithm, including following sub-step:
(1) rectangular area for needing to project is incorporated by origin of this vehicle;
(2) Multidimensional numerical dimensionality reduction;
(3) laser radar point is mapped to location of pixels;
(4) translational coordination origin;
(5) normalized coordinates value is to 0~255;
(6) spectrum mapping is done to grayscale image based on the depth information of preservation;
Each laser radar point cloud data can be attached to a reflectivity and receive natural conditions to characterize the point
The strength conditions of lower object reflection are the arrays that N row 4 arranges, first three is classified as the three-dimensional coordinate information of laser radar, and the
Four are classified as the Reflection intensity information of corresponding points.
The present invention will be schemed using image combination laser radar and its multi-modal lane scene detection method of depth information
Laser radar point cloud data is projected directly on image as pixel coordinate system and radar fix system translate rotation, overcomes biography
That unites is only insufficient to lane scene ability to express with image modalities, has real-time good and obstacle target detection is accurate
Advantage.
Detailed description of the invention
Fig. 1 is flow chart of the invention.
Specific embodiment
With reference to embodiment, the embodiment of the present invention is furthur described in detail.Following embodiment is used for
Illustrate the present invention, but is not intended to limit the scope of the invention.
A kind of unmanned vehicle object detection method based on multi-modal deep learning, comprising the following steps:
S1, the image data of monocular camera and the point cloud of laser radar are acquired using in-vehicle camera and mobile lidar
Data, and synchronize;
S2, basis look down visual angle projection algorithm process generation radar and look down perspective data, incorporate into first by origin of this vehicle
The rectangular area for needing to project, is then mapped to location of pixels for laser radar point.Each point cloud data point can be attached to one
Reflectivity receives the strength conditions that object reflects under natural conditions to characterize the point, is the array that N row 4 arranges.First three columns
The three-dimensional coordinate information of laser radar is saved, and the 4th column then save the Reflection intensity information of corresponding points.Standardize reflected intensity
It is worth 0-255, then grayscale image does spectrum mapping;
S3, laser radar top view further feature is extracted using convolutional neural networks, is operated by convolution operation, pondization
Feature extraction is carried out to the laser radar radar top view of input;
Laser radar top view further feature after S4, extraction is sent into RPN network, and mentioned feature is classified and returned
Return, obtain laser radar 2D looks down perspective data candidate frame.
S5, image further feature is extracted using convolutional neural networks, the figure to input is operated by convolution operation, pondization
As carrying out feature extraction;
Image further feature after S6, extraction is sent into RPN network, and mentioned feature is classified and returned, image is obtained
Data candidate frame, as shown in Fig. 1 right part of flg;
S7, the characteristic pattern that laser radar and the respective candidate frame of image are projected to same specification, by each coordinate first divided by
The picture of input obtains the frame coordinate on characteristic pattern with the ratio of the size of characteristic pattern, is obtained using the down-sampling pond that is averaged
Output, due to the scale that the image and radar of input look down visual angle figure be all it is unified, using average pond;
S8, removal redundancy candidate frame, the present invention is using laser radar candidate frame as standard, according to candidate frame central threshold model
It encloses, abandons the redundancy anchor in image.
S9, three dimensional detection frame is obtained, is classified by softmax function, the loss carried by tensorflow
Each coordinate points of function -- mean square error (MSE) Lai Huigui three dimensional detection frame.
Wherein, softmax function is the real vector of a N-dimensional to be mapped to another N-dimensional real vector, normalization
Whole elements between 0 to 1 to get the probability for belonging to each classification to each detection block,
Tensorflow is that a kind of deep learning frame of open source is similar with caffe, pytorch etc..
It is described that look down visual angle projection algorithm process as follows:
(1) rectangular area for needing to project is incorporated by origin of this vehicle;
Ff=np, logical_and ((x_lidar>0), (x_lidar<q))
Ss=np, logical_and ((y_lidar>-t), (y_lidar<t))
(2) Multidimensional numerical dimensionality reduction;
Indices=np.argwhere (np.logical_and (ff, ss)) .flatten ()
(3) laser radar point is mapped to location of pixels;
For y_points=-t ... .t do
For x_points=-0 ... .q do
X_img=(- y_pointaes) .astype (np.int32)
Y_img=(- x_pointaes) .astype (np.int32)
end for
(4) translational coordination origin;
X_img-=-t
Y_img-=0
(5) normalized coordinates value is to 0~255;
(6) spectrum mapping is done to grayscale image based on the depth information of preservation.
Wherein, np.logical_and () function is logic and function in (1), and x_lidar is the seat of x in radar points cloud
Mark, it is to set true which, which meets 0 < x_lidar < q value for ff corresponding position, is unsatisfactory for being set to false, y_lidar is same
Reason;(2) np.argwhere () function is index function in, finds the index that ff and ss same position is true, that is, finds out
Meet 0 < x_lidar < q in radar points cloud, the index of-t < y_lidar < t point, the index found out is a Multidimensional numerical, is needed
One-dimensional array is translated into flatten () function;(3) it converts point cloud data to for two-dimensional image data in,
X_img and y_img is the coordinate in top view, and res is the actual distance between two pixels, then passes through astype ()
Function is converted into integer value;(4) in that radar top view coordinate system and image coordinate system is unified.
A possibility that RPN prediction anchor in the present invention the is background or prospect i.e. process of classification, and extract anchor namely recurrence
No matter position of the anchor on characteristic pattern out, take the anchor of what ratio, last testing result it is accurate whether largely by
The candidate frame that RPN network provides, which is nominated, to be determined.The mistake nomination candidate frame even provided is more, and it is complicated not only to increase the algorithm time
Degree, also has a significant impact to accuracy.Anchor described in the text is the regional frame on characteristic pattern in fact.
The softmax function that the present invention uses is the real vector of a N-dimensional to be mapped to another N-dimensional real vector,
Whole elements are normalized between 0 to 1, more classification tasks are finally determined according to its range.
The advantage of present invention combination multisensor provides a kind of complexity that copes with and may face in unmanned vehicle traveling
Scene perception under scene has been evaded and has only detected bring lane object scene false retrieval, missing inspection situation with image modalities, so that
The result of target detection is relatively reliable.
And the present invention overcomes the shortcomings of that image modalities to lane scene ability to express, propose a kind of image combination laser
Radar utilizes the multi-modal vehicle object detection method of deep learning, by the fusion of multi-modal information, when realizing vehicle driving pair
In the accurate perception of scene around.
Loss function used in the present invention has quantified the gap between the predicted value of artificial neural network and true value,
And train the purpose of neural network to be that and reduce this gap, therefore, minimizes loss function and namely train neural network
Target.
Loss function of the present invention be intersect entropy function and image section loss function multinomial and form:
Wherein, Nl、NcRespectively represent the number of training of laser radar and image;ynFor the mark value of n-th of sample, it is
Vehicle is represented when 1, represents other objects when being 0,It is convolutional neural networks for the predicted value of sample;W is net
Parameter in network, multinomial first half are exactly the loss function definition about laser radar, and aft section is damaged about image
Considering for function is lost, image impairment function loses L by confidence levelconfL is lost with positioninglocIt forms, x represents prediction block in formula
Whether matched with true frame, c represents the matching degree of prediction block Yu true frame, and α is balance factor, when cross validation
Value is that 1, l represents prediction block, and g represents true frame, also represent the finally target detection classification level of confidence that returns out and
Whether the accurate positioning of each coordinate points of target detection, multiple risks function constraint is exactly to fit close to label data
True value improves the performance level of network entirety whereby.
The above is only a preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For member, without departing from the technical principles of the invention, several improvements and modifications can also be made, these improve and become
Type also should be regarded as protection scope of the present invention.
Claims (7)
1. a kind of unmanned vehicle object detection method based on multi-modal deep learning, which comprises the following steps:
(1) data acquire: utilizing the image data and laser radar of in-vehicle camera and mobile lidar acquisition monocular camera
Point cloud data;
(2) it handles laser radar point cloud data: looking down visual angle figure by look down that visual angle projection algorithm obtains laser radar;
(3) feature extraction: the further feature of laser radar top view and image data is extracted respectively using convolutional neural networks;
(4) tagsort returns, and respectively obtains the candidate frame of laser radar and the multiple targets of image data;
(5) it projects: laser radar being looked down into visual angle figure and the respective candidate frame of image data projects to the characteristic pattern of same specification;
(6) it removes redundancy candidate frame: using laser radar candidate frame as standard, according to candidate frame central threshold range, abandoning image
Redundancy anchor in data does not occur in laser radar such as but the anchor of candidate frame occurs in image data, then deletes;
(7) the nomination correction of image aspects avoids differing biggish laser radar target nomination with practical;
(8) three dimensional detection frame is exported: the laser radar point cloud data array of rule-based input convolutional network and the depth of reservation
Information is returned out together by loss function to be detected the coordinate of vehicle and saves.
2. a kind of unmanned vehicle object detection method based on multi-modal deep learning according to claim 1, feature exist
In, the loss function be intersect entropy function and image section loss function multinomial and:
Wherein, Nl、NcRespectively represent the number of training of laser radar and image, ynFor the mark value of n-th of sample,For volume
Product neural network is for the predicted value of sample, and W is the parameter in network, and multinomial first half is exactly the damage about laser radar
Function definition is lost, aft section is considering about image impairment function, and image impairment function loses L by confidence levelconfWith it is fixed
Bit-loss LlocComposition, x represent whether prediction block matches with true frame, and c represents the matching degree of prediction block Yu true frame, and α is flat
Weigh the factor, and l represents prediction block, and g represents true frame.
3. a kind of unmanned vehicle object detection method based on multi-modal deep learning according to claim 1, feature exist
In, further include data synchronizing step, by in-vehicle camera and mobile lidar acquisition monocular camera image data and laser thunder
The point cloud data reached synchronizes processing.
4. a kind of unmanned vehicle object detection method based on multi-modal deep learning according to claim 1, feature exist
In the coordinate quantity of the detected vehicle is 8.
5. a kind of unmanned vehicle object detection method based on multi-modal deep learning according to claim 1, feature exist
In the tagsort returns, the sub-step including following non-sequential execution:
A. the feature extracted from laser radar point cloud data is sent into RPN network, mentioned feature is classified and returned, is obtained
To the frame coordinate value of targets multiple under 3D coordinate, by converting the candidate frame for obtaining 2D and looking down under visual angle;
B. the extracted feature of institute it will be sent into RPN network from image data, mentioned feature is classified and returned, figure is obtained
As the candidate frame of targets multiple in plane.
6. a kind of unmanned vehicle object detection method based on multi-modal deep learning according to claim 1, feature exist
In described looks down visual angle projection algorithm, including following sub-step:
(1) rectangular area for needing to project is incorporated by origin of this vehicle;
(2) Multidimensional numerical dimensionality reduction;
(3) laser radar point is mapped to location of pixels;
(4) translational coordination origin;
(5) normalized coordinates value is to 0~255;
(6) spectrum mapping is done to grayscale image based on the depth information of preservation.
7. a kind of unmanned vehicle object detection method based on multi-modal deep learning according to claim 1, feature exist
In each laser radar point cloud data can be attached to a reflectivity and receive object under natural conditions to characterize the point
The strength conditions of reflection, are the arrays that N row 4 arranges, first three is classified as the three-dimensional coordinate information of laser radar, and the 4th is classified as pair
The Reflection intensity information that should be put.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811388553.3A CN109543601A (en) | 2018-11-21 | 2018-11-21 | A kind of unmanned vehicle object detection method based on multi-modal deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811388553.3A CN109543601A (en) | 2018-11-21 | 2018-11-21 | A kind of unmanned vehicle object detection method based on multi-modal deep learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109543601A true CN109543601A (en) | 2019-03-29 |
Family
ID=65848703
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811388553.3A Pending CN109543601A (en) | 2018-11-21 | 2018-11-21 | A kind of unmanned vehicle object detection method based on multi-modal deep learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109543601A (en) |
Cited By (25)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
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 |
CN110232315A (en) * | 2019-04-29 | 2019-09-13 | 华为技术有限公司 | Object detection method and device |
CN110441789A (en) * | 2019-06-27 | 2019-11-12 | 南京莱斯信息技术股份有限公司 | A kind of non-homogeneous thread laser radar Target Recognition Algorithms |
CN110531759A (en) * | 2019-08-02 | 2019-12-03 | 深圳大学 | Path generating method, device, computer equipment and storage medium are explored by robot |
CN110706288A (en) * | 2019-10-10 | 2020-01-17 | 上海眼控科技股份有限公司 | Target detection method, device, equipment and readable storage medium |
CN110743818A (en) * | 2019-11-29 | 2020-02-04 | 苏州嘉诺环境工程有限公司 | Garbage sorting system and garbage sorting method based on vision and deep learning |
CN110781927A (en) * | 2019-10-11 | 2020-02-11 | 苏州大学 | Target detection and classification method based on deep learning under cooperation of vehicle and road |
CN110992731A (en) * | 2019-12-12 | 2020-04-10 | 苏州智加科技有限公司 | Laser radar-based 3D vehicle detection method and device and storage medium |
CN111209825A (en) * | 2019-12-31 | 2020-05-29 | 武汉中海庭数据技术有限公司 | Method and device for dynamic target 3D detection |
CN111242015A (en) * | 2020-01-10 | 2020-06-05 | 同济大学 | Method for predicting driving danger scene based on motion contour semantic graph |
CN111274988A (en) * | 2020-02-10 | 2020-06-12 | 安徽大学 | Multispectral-based vehicle weight identification method and device |
CN111693050A (en) * | 2020-05-25 | 2020-09-22 | 电子科技大学 | Indoor medium and large robot navigation method based on building information model |
CN111695403A (en) * | 2020-04-19 | 2020-09-22 | 东风汽车股份有限公司 | 2D and 3D image synchronous detection method based on depth perception convolutional neural network |
WO2020199834A1 (en) * | 2019-04-03 | 2020-10-08 | 腾讯科技(深圳)有限公司 | Object detection method and apparatus, and network device and storage medium |
CN111860493A (en) * | 2020-06-12 | 2020-10-30 | 北京图森智途科技有限公司 | Target detection method and device based on point cloud data |
CN112560717A (en) * | 2020-12-21 | 2021-03-26 | 青岛科技大学 | Deep learning-based lane line detection method |
WO2021081808A1 (en) * | 2019-10-30 | 2021-05-06 | 深圳市大疆创新科技有限公司 | Artificial neural network-based object detection system and method |
CN113239749A (en) * | 2021-04-27 | 2021-08-10 | 四川大学 | Cross-domain point cloud semantic segmentation method based on multi-modal joint learning |
CN113516685A (en) * | 2021-07-09 | 2021-10-19 | 东软睿驰汽车技术(沈阳)有限公司 | Target tracking method, device, equipment and storage medium |
CN113780257A (en) * | 2021-11-12 | 2021-12-10 | 紫东信息科技(苏州)有限公司 | Multi-mode fusion weak supervision vehicle target detection method and system |
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 |
CN115019270A (en) * | 2022-05-31 | 2022-09-06 | 电子科技大学 | Automatic driving night target detection method based on sparse point cloud prior information |
CN115457496A (en) * | 2022-09-09 | 2022-12-09 | 北京百度网讯科技有限公司 | Automatic driving retaining wall detection method and device and vehicle |
CN115527074A (en) * | 2022-11-29 | 2022-12-27 | 深圳依时货拉拉科技有限公司 | Vehicle detection frame generation method and device and computer equipment |
CN117475397A (en) * | 2023-12-26 | 2024-01-30 | 安徽蔚来智驾科技有限公司 | Target annotation data acquisition method, medium and device based on multi-mode sensor |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP2637126A2 (en) * | 2012-03-07 | 2013-09-11 | Ricoh Company, Ltd. | Method and apparatus for detecting vehicle |
CN103793722A (en) * | 2014-03-07 | 2014-05-14 | 电子科技大学 | Rapid vehicle detection method and device under low resolution |
CN105701479A (en) * | 2016-02-26 | 2016-06-22 | 重庆邮电大学 | Intelligent vehicle multi-laser radar fusion recognition method based on target features |
WO2016113904A1 (en) * | 2015-01-16 | 2016-07-21 | 株式会社日立製作所 | Three-dimensional-information-calculating device, method for calculating three-dimensional information, and autonomous mobile device |
CN107451602A (en) * | 2017-07-06 | 2017-12-08 | 浙江工业大学 | A kind of fruits and vegetables detection method based on deep learning |
WO2018056212A1 (en) * | 2016-09-22 | 2018-03-29 | 株式会社デンソー | Object detection device and object detection method |
WO2018170472A1 (en) * | 2017-03-17 | 2018-09-20 | Honda Motor Co., Ltd. | Joint 3d object detection and orientation estimation via multimodal fusion |
WO2018204128A1 (en) * | 2017-05-01 | 2018-11-08 | Mentor Graphics Development (Deutschland) Gmbh | Embedded automotive perception with machine learning classification of sensor data |
-
2018
- 2018-11-21 CN CN201811388553.3A patent/CN109543601A/en active Pending
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP2637126A2 (en) * | 2012-03-07 | 2013-09-11 | Ricoh Company, Ltd. | Method and apparatus for detecting vehicle |
CN103793722A (en) * | 2014-03-07 | 2014-05-14 | 电子科技大学 | Rapid vehicle detection method and device under low resolution |
WO2016113904A1 (en) * | 2015-01-16 | 2016-07-21 | 株式会社日立製作所 | Three-dimensional-information-calculating device, method for calculating three-dimensional information, and autonomous mobile device |
CN105701479A (en) * | 2016-02-26 | 2016-06-22 | 重庆邮电大学 | Intelligent vehicle multi-laser radar fusion recognition method based on target features |
WO2018056212A1 (en) * | 2016-09-22 | 2018-03-29 | 株式会社デンソー | Object detection device and object detection method |
WO2018170472A1 (en) * | 2017-03-17 | 2018-09-20 | Honda Motor Co., Ltd. | Joint 3d object detection and orientation estimation via multimodal fusion |
WO2018204128A1 (en) * | 2017-05-01 | 2018-11-08 | Mentor Graphics Development (Deutschland) Gmbh | Embedded automotive perception with machine learning classification of sensor data |
CN107451602A (en) * | 2017-07-06 | 2017-12-08 | 浙江工业大学 | A kind of fruits and vegetables detection method based on deep learning |
Non-Patent Citations (1)
Title |
---|
林子彧: "多模态深度学习目标检测及其应用", 《中国优秀硕士学位论文全文数据库 (信息科技辑)》 * |
Cited By (36)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020199834A1 (en) * | 2019-04-03 | 2020-10-08 | 腾讯科技(深圳)有限公司 | Object detection method and apparatus, and network device and storage medium |
CN110232315A (en) * | 2019-04-29 | 2019-09-13 | 华为技术有限公司 | Object detection method and device |
CN110222626B (en) * | 2019-06-03 | 2021-05-28 | 宁波智能装备研究院有限公司 | Unmanned scene point cloud target labeling method based on deep learning algorithm |
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 |
CN110441789A (en) * | 2019-06-27 | 2019-11-12 | 南京莱斯信息技术股份有限公司 | A kind of non-homogeneous thread laser radar Target Recognition Algorithms |
CN110531759A (en) * | 2019-08-02 | 2019-12-03 | 深圳大学 | Path generating method, device, computer equipment and storage medium are explored by robot |
CN110706288A (en) * | 2019-10-10 | 2020-01-17 | 上海眼控科技股份有限公司 | Target detection method, device, equipment and readable storage medium |
CN110781927A (en) * | 2019-10-11 | 2020-02-11 | 苏州大学 | Target detection and classification method based on deep learning under cooperation of vehicle and road |
CN110781927B (en) * | 2019-10-11 | 2023-05-23 | 苏州大学 | Target detection and classification method based on deep learning under vehicle-road cooperation |
WO2021081808A1 (en) * | 2019-10-30 | 2021-05-06 | 深圳市大疆创新科技有限公司 | Artificial neural network-based object detection system and method |
CN110743818A (en) * | 2019-11-29 | 2020-02-04 | 苏州嘉诺环境工程有限公司 | Garbage sorting system and garbage sorting method based on vision and deep learning |
CN110992731A (en) * | 2019-12-12 | 2020-04-10 | 苏州智加科技有限公司 | Laser radar-based 3D vehicle detection method and device and storage medium |
CN111209825A (en) * | 2019-12-31 | 2020-05-29 | 武汉中海庭数据技术有限公司 | Method and device for dynamic target 3D detection |
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 |
CN111242015B (en) * | 2020-01-10 | 2023-05-02 | 同济大学 | Method for predicting driving dangerous scene based on motion profile semantic graph |
CN111242015A (en) * | 2020-01-10 | 2020-06-05 | 同济大学 | Method for predicting driving danger scene based on motion contour semantic graph |
CN111274988A (en) * | 2020-02-10 | 2020-06-12 | 安徽大学 | Multispectral-based vehicle weight identification method and device |
CN111274988B (en) * | 2020-02-10 | 2023-03-24 | 安徽大学 | Multispectral-based vehicle weight identification method and device |
CN111695403B (en) * | 2020-04-19 | 2024-03-22 | 东风汽车股份有限公司 | Depth perception convolutional neural network-based 2D and 3D image synchronous detection method |
CN111695403A (en) * | 2020-04-19 | 2020-09-22 | 东风汽车股份有限公司 | 2D and 3D image synchronous detection method based on depth perception convolutional neural network |
CN111693050A (en) * | 2020-05-25 | 2020-09-22 | 电子科技大学 | Indoor medium and large robot navigation method based on building information model |
CN111860493B (en) * | 2020-06-12 | 2024-02-09 | 北京图森智途科技有限公司 | Target detection method and device based on point cloud data |
CN111860493A (en) * | 2020-06-12 | 2020-10-30 | 北京图森智途科技有限公司 | Target detection method and device based on point cloud data |
CN112560717B (en) * | 2020-12-21 | 2023-04-21 | 青岛科技大学 | Lane line detection method based on deep learning |
CN112560717A (en) * | 2020-12-21 | 2021-03-26 | 青岛科技大学 | Deep learning-based lane line detection method |
CN113239749A (en) * | 2021-04-27 | 2021-08-10 | 四川大学 | Cross-domain point cloud semantic segmentation method based on multi-modal joint learning |
CN113516685A (en) * | 2021-07-09 | 2021-10-19 | 东软睿驰汽车技术(沈阳)有限公司 | Target tracking method, device, equipment and storage medium |
CN113780257A (en) * | 2021-11-12 | 2021-12-10 | 紫东信息科技(苏州)有限公司 | Multi-mode fusion weak supervision vehicle target detection method and system |
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 |
CN115019270A (en) * | 2022-05-31 | 2022-09-06 | 电子科技大学 | Automatic driving night target detection method based on sparse point cloud prior information |
CN115019270B (en) * | 2022-05-31 | 2024-04-19 | 电子科技大学 | Automatic driving night target detection method based on sparse point cloud priori information |
CN115457496A (en) * | 2022-09-09 | 2022-12-09 | 北京百度网讯科技有限公司 | Automatic driving retaining wall detection method and device and vehicle |
CN115457496B (en) * | 2022-09-09 | 2023-12-08 | 北京百度网讯科技有限公司 | Automatic driving retaining wall detection method and device and vehicle |
CN115527074A (en) * | 2022-11-29 | 2022-12-27 | 深圳依时货拉拉科技有限公司 | Vehicle detection frame generation method and device and computer equipment |
CN117475397A (en) * | 2023-12-26 | 2024-01-30 | 安徽蔚来智驾科技有限公司 | Target annotation data acquisition method, medium and device based on multi-mode sensor |
CN117475397B (en) * | 2023-12-26 | 2024-03-22 | 安徽蔚来智驾科技有限公司 | Target annotation data acquisition method, medium and device based on multi-mode sensor |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109543601A (en) | A kind of unmanned vehicle object detection method based on multi-modal deep learning | |
Xia et al. | Geometric primitives in LiDAR point clouds: A review | |
CN110415342B (en) | Three-dimensional point cloud reconstruction device and method based on multi-fusion sensor | |
Maddern et al. | 1 year, 1000 km: The oxford robotcar dataset | |
Geiger et al. | Vision meets robotics: The kitti dataset | |
CN113111887B (en) | Semantic segmentation method and system based on information fusion of camera and laser radar | |
Zhou et al. | Self‐supervised learning to visually detect terrain surfaces for autonomous robots operating in forested terrain | |
CN110009765A (en) | A kind of automatic driving vehicle contextual data system and scene format method for transformation | |
CN114254696A (en) | Visible light, infrared and radar fusion target detection method based on deep learning | |
Zhao et al. | Road network extraction from airborne LiDAR data using scene context | |
CN113989797A (en) | Three-dimensional dynamic target detection method and device based on voxel point cloud fusion | |
EP4377913A1 (en) | Training method for training a change detection system, training set generating method therefor, and change detection system | |
US20220410381A1 (en) | Systems and methods for picking objects using 3-d geometry and segmentation | |
CN117058646B (en) | Complex road target detection method based on multi-mode fusion aerial view | |
CN114639115A (en) | 3D pedestrian detection method based on fusion of human body key points and laser radar | |
CN114298151A (en) | 3D target detection method based on point cloud data and image data fusion | |
CN116612468A (en) | Three-dimensional target detection method based on multi-mode fusion and depth attention mechanism | |
CN116664856A (en) | Three-dimensional target detection method, system and storage medium based on point cloud-image multi-cross mixing | |
Jindal et al. | Bollard segmentation and position estimation from lidar point cloud for autonomous mooring | |
CN114494248A (en) | Three-dimensional target detection system and method based on point cloud and images under different visual angles | |
Zhang et al. | Infrastructure 3D Target detection based on multi-mode fusion for intelligent and connected vehicles | |
CN117372991A (en) | Automatic driving method and system based on multi-view multi-mode fusion | |
CN116817891A (en) | Real-time multi-mode sensing high-precision map construction method | |
Lertniphonphan et al. | 2d to 3d label propagation for object detection in point cloud | |
CN116740514A (en) | Space-time error tolerant multi-agent cooperative sensing method and device and electronic equipment |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190329 |
|
RJ01 | Rejection of invention patent application after publication |