CN106780605A - A kind of detection method of the object crawl position based on deep learning robot - Google Patents
A kind of detection method of the object crawl position based on deep learning robot Download PDFInfo
- Publication number
- CN106780605A CN106780605A CN201611181461.9A CN201611181461A CN106780605A CN 106780605 A CN106780605 A CN 106780605A CN 201611181461 A CN201611181461 A CN 201611181461A CN 106780605 A CN106780605 A CN 106780605A
- Authority
- CN
- China
- Prior art keywords
- capture area
- input
- crawl position
- candidate capture
- deep learning
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Image Analysis (AREA)
Abstract
The present invention is applied to robot and captures field, there is provided a kind of detection method of the object crawl position based on deep learning robot, the method comprises the following steps:RGB D images comprising object are gathered by sensor;Candidate capture area is divided in the target area of RGB D images;Keep the length-width ratio of candidate capture area constant, the size of candidate capture area is amplified to the size of neutral net input requirements;Input vector is built to the candidate capture area after amplification;Whitening processing is carried out to input vector, the input vector after the whitening processing is input to the neutral net for training;The score of each candidate capture area is obtained, the candidate capture area of highest scoring is defined as crawl position.The RGB D images of the object by obtaining, you can determine the crawl position of the object, robot by the crawl position can realize it is any realize the crawl of object, and do not need artificial participation.
Description
Technical field
The invention belongs to robot crawl field, more particularly to a kind of object crawl position based on deep learning robot
The detection method put.
Background technology
In order to nurse the handicapped personage such as the elderly, disabled person, to familiar object in home environment, such as teacup beverage
The crawl of bottle, books etc., as the indispensable critical function demand of home-services robot.Different from industrial robot in knot
To the crawl of workpiece under structure environment, intelligent grabbing of the service robot under home environment is faced with lot of challenges, for example, move
State environment, illumination variation, tens or even hundreds of target object, mutually blocking between complex background, object.
At present, robot crawl detection technique includes following several:The crawl feature of engineer's object, by target
Crawl feature sets up crawl model, detects crawl position, and the method for the crawl feature of existing engineer's object both took
Substantial amounts of artificial participation is needed again, and cannot accurately detect crawl position for the unseen object of robot, it is impossible to
Perform grasping movement.
The content of the invention
The embodiment of the present invention provides a kind of detection method of the object crawl position based on deep learning robot, it is intended to
The method for solving the crawl feature of existing engineer's object, not only took but also needed substantial amounts of artificial participation, and for
The unseen object of robot cannot accurately detect crawl position, it is impossible to perform grasping movement problem.
The present invention is achieved in that a kind of detection method of the object crawl position based on deep learning robot,
Methods described comprises the following steps:
S1. the RGB-D images comprising object are gathered by sensor;
S2. candidate capture area is divided in the target area of RGB-D images;
S3. keep the length-width ratio of the candidate capture area constant, the size of the candidate capture area is amplified to god
Through the size of network inputs requirement;
S4. input vector is built to the candidate capture area after the amplification;
S5. whitening processing is carried out to the input vector, the input vector after the whitening processing is input to and is trained
Neutral net;
S6. the score of each candidate capture area is obtained, the candidate capture area of the highest scoring is defined as crawl
Position.
The embodiment of the present invention divides candidate and grabs by obtaining the RGB-D images of object, to the target area of RGB-D images
Take region, and be amplified to the size of neutral net input requirements, input vector is built to the candidate capture area after amplification, by structure
The input vector input neutral net built up, obtains the score of a candidate capture area, and the candidate capture area of highest scoring is true
It is set to the crawl position of object, the RGB-D images of the object by obtaining, you can determine the crawl position of the object,
Robot by the crawl position can realize it is any realize the crawl of object, and do not need artificial participation.
Brief description of the drawings
Fig. 1 is the stream of the object crawl position detection method based on deep learning robot provided in an embodiment of the present invention
Cheng Tu.
Specific embodiment
In order to make the purpose , technical scheme and advantage of the present invention be clearer, it is right below in conjunction with drawings and Examples
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.
Fig. 1 is the stream of the object crawl position detection method based on deep learning robot provided in an embodiment of the present invention
Cheng Tu, the method comprises the following steps:
S1. the RGB-D images comprising object are gathered by sensor;
The embodiment of the present invention obtains the high-resolution RGB image and depth of crawl object using Microsoft's Kinect sensor
Degree image, RGB image contains crawl target surface colouring information and texture information, and depth image contains crawl object
Spatial form information, each pixel value in depth image illustrate sensor to crawl object distance, RGB image and
Pixel between depth image is one-to-one, constitutes RGB-D images.
S2. candidate capture area is divided in the target area of RGB-D images;
In embodiments of the present invention, the target area of RGB-D images is extracted using background subtraction, is set in target area
Determine sliding window, candidate capture area is extracted by the movement of sliding window, the size of the active window is candidate's crawl
The size in region, the embodiment of the present invention uses baxter tow-armed robots, and the end effector of robot is clamping jaw, sliding window
It is rectangular slide window, the size of sliding window is the size according to gripper to be determined, active window is set to 30 pixel × 10
The rectangular active window of pixel, therefore the size of candidate capture area is candidate's crawl rectangle of 30 pixel × 10 pixels.
S3. keep the length-width ratio of candidate capture area constant, the size of candidate capture area is amplified to neutral net defeated
Enter the size of requirement;
In embodiments of the present invention, the requirement according to neutral net to input sample size, is keeping candidate capture area
Length-width ratio it is constant in the case of, filled by 0 value, or the size of candidate capture area is amplified to neutral net by border extended
The size of input requirements, in embodiments of the present invention, requirement of the neutral net to input sample size is 32 pixel × 32 pixels.
S4. input vector is built to the candidate capture area after amplification;
In embodiments of the present invention, to candidate capture area build 7 input vectors of passage, 7 inputs of passage to
Amount includes:Surface normal of the depth data on three directions of x, y, z axle, yuv data is obtained from depth image to be converted into
On tri- passages of Y, U, V vector and depth image is converted into vector.
In embodiments of the present invention, if step S3 is to fill for the size of candidate capture area to be amplified to nerve by 0 value
The size of network inputs requirement, due to having substantial amounts of 0 value, and different candidate capture areas in the input vector after filling
Input vector in fill 0 be worth quantity different, can finally influence the output score of neutral net, therefore, filled out to eliminate
The influence of 0 value filled to waiting the element in input vector, it is necessary to be multiplied by a zoom factor, the value to element in input vector is entered
Row scaling, the computing formula of zoom factor is as follows:
Wherein,It is i-th zoom factor of element in t-th input vector of sample,For passage r amplification because
Son, when i-th element x in input vectoriWhen belonging to passage r, Sr,iValue be 1, otherwise Sr,iValue be 0, work as xthiIt is not 0
During Filling power,Value be 1, otherwiseValue be 0.
Additionally, as the preferred embodiments of the present invention, it is contemplated that zoom factor crosses conference causes the distortion of input data, contracting
Put the factor and be up to certain value C, i.e.,The value of C is 4
S5. whitening processing is carried out to the input vector for building, the input vector after whitening processing is input to what is trained
Neutral net;
In embodiments of the present invention, in order to reduce the redundancy of input, the input vector to building carries out whitening processing,
The process of whitening processing includes:The input vector for being input into each passage is subtracted into respective average value, then divided by by 7 passages
The standard deviation of the mix vector of input vector composition.
S6. the score of each candidate capture area is obtained, the candidate region of highest scoring is defined as crawl position.
The embodiment of the present invention divides candidate and grabs by obtaining the RGB-D images of object, to the target area of RGB-D images
Take region, and be amplified to the size of neutral net input requirements, input vector is built to the candidate capture area after amplification, by structure
The input vector input neutral net built up, obtains the score of a candidate capture area, and the candidate capture area of highest scoring is true
It is set to the crawl position of object, the RGB-D images of the object by obtaining, you can determine the crawl position of the object,
Robot by the crawl position can realize it is any realize the crawl of object, and do not need artificial participation.
In embodiments of the present invention, also included before the step S1:
S7. neutral net is built;
In embodiments of the present invention, the neutral net of structure is sparse certainly by 7168 neuron input layers, 200 neurons
Encoder and sigmoid output layers are constituted.
S8. off-line training is carried out to the neutral net for building.
In embodiments of the present invention, by giving the input and output of sample, W when obtaining optimal by training, then use W
Calculate the prediction output of given input.
In embodiments of the present invention, off-line training is carried out to the neutral net for building and specifically includes following steps:
S81. the sample for giving is pre-processed using step S1-S5;
S82. the given sample input neutral net that will have been pre-processed, and the output result of sample is given, using unsupervised
Training iteration 200 times, trains 2 sparse self-encoding encoders of hidden layer to initialize hidden layer weights;
Initialization hidden layer weights W of the sparse self-encoding encoder when cost function is optimal*Formula is as follows:
Wherein,It is input vector x(t)Reconstruction, g (h) be openness penalty, λ is openness penalty
Coefficient, f (W) is Regularization function, and β is the coefficient of Regularization function,For t-th i-th of input sample input vector yuan
Element, Wi,jIt is weights of i-th element on j-th hidden neuron,It is that t-th input sample is vectorial in j-th hidden layer
Output on neuron, σ is sigmoid activation primitives, W*Sparse self-encoding encoder initializes hidden layer when being optimal cost function
Weights.
Sparse self-encoding encoder in the embodiment of the present invention includes the Sparse self-encoding encoder and the second layer of ground floor hidden layer
The standardized sparse self-encoding encoder of hidden layer;
When the sparse self-encoding encoder is the first hidden layer Sparse self-encoding encoder, the regularization combined using L2 and L1
Method, regular function isWherein | | W | |1It is the corresponding Regularization functions of L1,It is L2 pairs
The Regularization function answered, the regularization coefficient ε of L12=0.0003, L2 regularization coefficient ε1=0.001, wherein to f1(W) add
Slight bias amount 0.00001;0 worth interference in input vector is wherein avoided by adding the method for slight bias, this layer dilute
It is 3 to dredge property penalty coefficient, and the output of Sparse self-encoding encoder is the real number between 0 to 1;
When the sparse self-encoding encoder is the second hidden layer standardized sparse self-encoding encoder, using L1 regularization methods, canonical
Function is f2(W)=ε2||W||1, L1 regularization coefficients, ε2=0.0003, this layer of openness penalty coefficient is 3, and standardized sparse is certainly
The output of encoder is 0 or 1.
S83. the sample that will have been pre-processed passes through back-propagation algorithm Training iteration 10 times, to whole network parameter
Carry out global optimization.
Presently preferred embodiments of the present invention is the foregoing is only, is not intended to limit the invention, it is all in essence of the invention
Any modification, equivalent and improvement made within god and principle etc., should be included within the scope of the present invention.
Claims (9)
1. a kind of detection method of the object crawl position based on deep learning robot, it is characterised in that methods described bag
Include following steps:
S1. the RGB-D images comprising object are gathered by sensor;
S2. candidate capture area is divided in the target area of the RGB-D images;
S3. keep the length-width ratio of the candidate capture area constant, the size of the candidate capture area is amplified to nerve net
The size of network input requirements;
S4. input vector is built to the candidate capture area after the amplification;
S5. whitening processing is carried out to the input vector, the input vector after the whitening processing is input to the god for training
Through network;
S6. the score of each candidate capture area is obtained, the candidate capture area of the highest scoring is defined as crawl position.
2. the method for inspection of the object crawl position of deep learning robot is based on as claimed in claim 1, and its feature exists
In the candidate capture area is that the sliding window by setting moves to extract in the target area of the RGB-D images
's;
The size of the active window is the size of the candidate capture area.
3. the method for inspection of the object crawl position of deep learning robot is based on as claimed in claim 1, and its feature exists
In being filled by 0 value or the size of the candidate capture area be amplified to border extended the size of neutral net input requirements.
4. the detection method of the object crawl position of deep learning robot is based on as claimed in claim 3, and its feature exists
In when the size of the candidate capture area being amplified into the size of neutral net input requirements by 0 value filling, in step
Also include after rapid S4:
Element in the input vector is multiplied by a zoom factor, the value to element in the input vector is zoomed in and out;
The computing formula of the zoom factor is as follows:
Wherein,It is i-th zoom factor of element in t-th input vector of sample,It is the amplification factor of passage r,
When i-th element x in input vectoriWhen belonging to passage r, Sr,iValue be 1, otherwise Sr,iValue be 0, work as xthiIt is not 0 filling
During value,Value be 1, otherwiseValue be 0.
5. the detection method of the object crawl position of deep learning robot is based on as claimed in claim 4, and its feature exists
In the zoom factorThe value of C is 4.
6. the detection method of the object crawl position of deep learning robot is based on as claimed in claim 1, and its feature exists
In also including before step S1:
S7. neutral net is built;
S8. off-line training is carried out to the neutral net for building.
The neutral net is by 7168 neuron input layers, 200 sparse self-encoding encoders of neuron and sigmoid output layer groups
Into.
7. the detection method of the object crawl position of deep learning robot is based on as claimed in claim 6, and its feature exists
In the step S8 specifically includes following steps:
S81. the sample for giving is pre-processed using step S1-S5;
S82. the given sample that has pre-processed is input into neutral net, and given sample output result, using unsupervised
Training iteration 200 times, trains 2 sparse self-encoding encoders of hidden layer to initialize hidden layer weights;
S83. the sample for having pre-processed is passed through into back-propagation algorithm Training iteration 10 times, to whole network parameter
Carry out global optimization.
8. the detection method of the object crawl position of deep learning robot is based on as claimed in claim 7, and its feature exists
In initialization hidden layer weights W of the sparse self-encoding encoder when cost function is optimal*Computing formula is as follows:
Wherein,It is input vector x(t)Reconstruction, g (h) be openness penalty, λ is the coefficient of openness penalty, f
(W) it is Regularization function, β is the coefficient of Regularization function,It is t-th i-th element of input sample input vector, Wi,j
It is weights of i-th element on j-th hidden neuron,It is that t-th input sample is vectorial in j-th hidden neuron
On output, σ be sigmoid activation primitives.
9. the detection method of the object crawl position of deep learning robot is based on as claimed in claim 8, and its feature exists
In when the sparse self-encoding encoder is the first hidden layer Sparse self-encoding encoder, the Sparse self-encoding encoder uses L2
The regularization method combined with L1, regular function isWherein | | W | |1It is the corresponding canonicals of L1
Change function,It is the corresponding Regularization functions of L2, the regularization coefficient ε of L12=0.0003, L2 regularization coefficient ε1=
0.001, wherein to f1(W) addition slight bias amount 0.00001, openness the punishing of the first hidden layer Sparse self-encoding encoder
The coefficient lambda of penalty function1=3;
When the sparse self-encoding encoder is the second hidden layer standardized sparse self-encoding encoder, the standardized sparse self-encoding encoder uses L1
Regularization method, regular function is f2(W)=ε2||W||1, L1 regularization coefficients ε2=0.0003, the second hidden layer standard is dilute
Dredge the coefficient lambda of the openness penalty of self-encoding encoder2=3.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611181461.9A CN106780605A (en) | 2016-12-20 | 2016-12-20 | A kind of detection method of the object crawl position based on deep learning robot |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611181461.9A CN106780605A (en) | 2016-12-20 | 2016-12-20 | A kind of detection method of the object crawl position based on deep learning robot |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106780605A true CN106780605A (en) | 2017-05-31 |
Family
ID=58890864
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201611181461.9A Pending CN106780605A (en) | 2016-12-20 | 2016-12-20 | A kind of detection method of the object crawl position based on deep learning robot |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106780605A (en) |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107479501A (en) * | 2017-09-28 | 2017-12-15 | 广州智能装备研究院有限公司 | 3D parts suction methods based on deep learning |
CN107679477A (en) * | 2017-09-27 | 2018-02-09 | 深圳市未来媒体技术研究院 | Face depth and surface normal Forecasting Methodology based on empty convolutional neural networks |
CN108126914A (en) * | 2017-11-24 | 2018-06-08 | 上海发那科机器人有限公司 | More object robots method for sorting at random in a kind of material frame based on deep learning |
CN108280856A (en) * | 2018-02-09 | 2018-07-13 | 哈尔滨工业大学 | The unknown object that network model is inputted based on mixed information captures position and orientation estimation method |
CN108805004A (en) * | 2018-04-12 | 2018-11-13 | 深圳市商汤科技有限公司 | Functional area detection method and device, electronic equipment, storage medium, program |
CN108908334A (en) * | 2018-07-20 | 2018-11-30 | 汕头大学 | A kind of intelligent grabbing system and method based on deep learning |
CN109508707A (en) * | 2019-01-08 | 2019-03-22 | 中国科学院自动化研究所 | The crawl point acquisition methods of robot stabilized crawl object based on monocular vision |
CN109531584A (en) * | 2019-01-31 | 2019-03-29 | 北京无线电测量研究所 | A kind of Mechanical arm control method and device based on deep learning |
CN110208211A (en) * | 2019-07-03 | 2019-09-06 | 南京林业大学 | A kind of near infrared spectrum noise-reduction method for Detecting Pesticide |
CN110691676A (en) * | 2017-06-19 | 2020-01-14 | 谷歌有限责任公司 | Robot crawling prediction using neural networks and geometrically-aware object representations |
CN111310637A (en) * | 2020-02-11 | 2020-06-19 | 山西大学 | Robot target grabbing detection method based on scale invariant network |
CN111324095A (en) * | 2020-02-27 | 2020-06-23 | 金陵科技学院 | Unmanned shipment system of dry bulk material intelligent industrial robot |
CN111428731A (en) * | 2019-04-04 | 2020-07-17 | 深圳市联合视觉创新科技有限公司 | Multi-class target identification and positioning method, device and equipment based on machine vision |
JP2021517681A (en) * | 2018-12-12 | 2021-07-26 | 達闥机器人有限公司 | How to detect the target object gripping position of the robot |
CN116945210A (en) * | 2023-07-12 | 2023-10-27 | 深圳市永顺创能技术有限公司 | Robot intelligent control system based on machine vision |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105598965A (en) * | 2015-11-26 | 2016-05-25 | 哈尔滨工业大学 | Robot under-actuated hand autonomous grasping method based on stereoscopic vision |
CN105718959A (en) * | 2016-01-27 | 2016-06-29 | 中国石油大学(华东) | Object identification method based on own coding |
CN106094516A (en) * | 2016-06-08 | 2016-11-09 | 南京大学 | A kind of robot self-adapting grasping method based on deeply study |
-
2016
- 2016-12-20 CN CN201611181461.9A patent/CN106780605A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105598965A (en) * | 2015-11-26 | 2016-05-25 | 哈尔滨工业大学 | Robot under-actuated hand autonomous grasping method based on stereoscopic vision |
CN105718959A (en) * | 2016-01-27 | 2016-06-29 | 中国石油大学(华东) | Object identification method based on own coding |
CN106094516A (en) * | 2016-06-08 | 2016-11-09 | 南京大学 | A kind of robot self-adapting grasping method based on deeply study |
Non-Patent Citations (1)
Title |
---|
IAN LENZ 等: ""Deep Learning for Detecting Robotic Grasps"", 《百度学术》 * |
Cited By (24)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110691676A (en) * | 2017-06-19 | 2020-01-14 | 谷歌有限责任公司 | Robot crawling prediction using neural networks and geometrically-aware object representations |
US11554483B2 (en) | 2017-06-19 | 2023-01-17 | Google Llc | Robotic grasping prediction using neural networks and geometry aware object representation |
CN107679477A (en) * | 2017-09-27 | 2018-02-09 | 深圳市未来媒体技术研究院 | Face depth and surface normal Forecasting Methodology based on empty convolutional neural networks |
CN107479501A (en) * | 2017-09-28 | 2017-12-15 | 广州智能装备研究院有限公司 | 3D parts suction methods based on deep learning |
CN108126914A (en) * | 2017-11-24 | 2018-06-08 | 上海发那科机器人有限公司 | More object robots method for sorting at random in a kind of material frame based on deep learning |
CN108280856A (en) * | 2018-02-09 | 2018-07-13 | 哈尔滨工业大学 | The unknown object that network model is inputted based on mixed information captures position and orientation estimation method |
CN108280856B (en) * | 2018-02-09 | 2021-05-07 | 哈尔滨工业大学 | Unknown object grabbing pose estimation method based on mixed information input network model |
CN108805004A (en) * | 2018-04-12 | 2018-11-13 | 深圳市商汤科技有限公司 | Functional area detection method and device, electronic equipment, storage medium, program |
CN108805004B (en) * | 2018-04-12 | 2021-09-14 | 深圳市商汤科技有限公司 | Functional area detection method and device, electronic equipment and storage medium |
CN108908334A (en) * | 2018-07-20 | 2018-11-30 | 汕头大学 | A kind of intelligent grabbing system and method based on deep learning |
US11878433B2 (en) | 2018-12-12 | 2024-01-23 | Cloudminds Robotics Co., Ltd. | Method for detecting grasping position of robot in grasping object |
JP7085726B2 (en) | 2018-12-12 | 2022-06-17 | 達闥機器人股▲分▼有限公司 | How to detect the target object gripping position of the robot |
JP2021517681A (en) * | 2018-12-12 | 2021-07-26 | 達闥机器人有限公司 | How to detect the target object gripping position of the robot |
CN109508707A (en) * | 2019-01-08 | 2019-03-22 | 中国科学院自动化研究所 | The crawl point acquisition methods of robot stabilized crawl object based on monocular vision |
CN109531584A (en) * | 2019-01-31 | 2019-03-29 | 北京无线电测量研究所 | A kind of Mechanical arm control method and device based on deep learning |
CN111428731A (en) * | 2019-04-04 | 2020-07-17 | 深圳市联合视觉创新科技有限公司 | Multi-class target identification and positioning method, device and equipment based on machine vision |
CN111428731B (en) * | 2019-04-04 | 2023-09-26 | 深圳市联合视觉创新科技有限公司 | Multi-category identification positioning method, device and equipment based on machine vision |
CN110208211B (en) * | 2019-07-03 | 2021-10-22 | 南京林业大学 | Near infrared spectrum noise reduction method for pesticide residue detection |
CN110208211A (en) * | 2019-07-03 | 2019-09-06 | 南京林业大学 | A kind of near infrared spectrum noise-reduction method for Detecting Pesticide |
CN111310637B (en) * | 2020-02-11 | 2022-11-11 | 山西大学 | Robot target grabbing detection method based on scale invariant network |
CN111310637A (en) * | 2020-02-11 | 2020-06-19 | 山西大学 | Robot target grabbing detection method based on scale invariant network |
CN111324095A (en) * | 2020-02-27 | 2020-06-23 | 金陵科技学院 | Unmanned shipment system of dry bulk material intelligent industrial robot |
CN116945210A (en) * | 2023-07-12 | 2023-10-27 | 深圳市永顺创能技术有限公司 | Robot intelligent control system based on machine vision |
CN116945210B (en) * | 2023-07-12 | 2024-03-15 | 深圳市永顺创能技术有限公司 | Robot intelligent control system based on machine vision |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106780605A (en) | A kind of detection method of the object crawl position based on deep learning robot | |
CN107886069A (en) | A kind of multiple target human body 2D gesture real-time detection systems and detection method | |
CN108280856B (en) | Unknown object grabbing pose estimation method based on mixed information input network model | |
CN106097322B (en) | A kind of vision system calibration method based on neural network | |
CN104978580B (en) | A kind of insulator recognition methods for unmanned plane inspection transmission line of electricity | |
CN106951923B (en) | Robot three-dimensional shape recognition method based on multi-view information fusion | |
Browne et al. | Convolutional neural networks for image processing: an application in robot vision | |
CN105447529A (en) | Costume detection and attribute value identification method and system | |
CN110509273B (en) | Robot manipulator detection and grabbing method based on visual deep learning features | |
Shinzato et al. | Fast visual road recognition and horizon detection using multiple artificial neural networks | |
CN105205453A (en) | Depth-auto-encoder-based human eye detection and positioning method | |
CN104392228A (en) | Unmanned aerial vehicle image target class detection method based on conditional random field model | |
He et al. | Integrated moment-based LGMD and deep reinforcement learning for UAV obstacle avoidance | |
Shinzato et al. | A road following approach using artificial neural networks combinations | |
CN107688856A (en) | Indoor Robot scene active identification method based on deeply study | |
CN107967474A (en) | A kind of sea-surface target conspicuousness detection method based on convolutional neural networks | |
CN107424161A (en) | A kind of indoor scene image layout method of estimation by thick extremely essence | |
CN106780546A (en) | The personal identification method of the motion blur encoded point based on convolutional neural networks | |
Saxena et al. | Garment recognition and grasping point detection for clothing assistance task using deep learning | |
Zhang et al. | Learn to navigate maplessly with varied LiDAR configurations: A support point-based approach | |
CN107414830A (en) | A kind of carrying machine human arm manipulation multi-level mapping intelligent control method and system | |
Komer et al. | BatSLAM: Neuromorphic spatial reasoning in 3D environments | |
Shinzato et al. | Path recognition for outdoor navigation using artificial neural networks: Case study | |
Guo et al. | Learning pushing skills using object detection and deep reinforcement learning | |
Zhao et al. | Multi-purpose tactile perception based on deep learning in a new tendon-driven optical tactile sensor |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170531 |