CN109635750A - A kind of compound convolutional neural networks images of gestures recognition methods under complex background - Google Patents
A kind of compound convolutional neural networks images of gestures recognition methods under complex background Download PDFInfo
- Publication number
- CN109635750A CN109635750A CN201811534388.8A CN201811534388A CN109635750A CN 109635750 A CN109635750 A CN 109635750A CN 201811534388 A CN201811534388 A CN 201811534388A CN 109635750 A CN109635750 A CN 109635750A
- Authority
- CN
- China
- Prior art keywords
- images
- gestures
- gesture
- convolutional neural
- neural networks
- 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
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
- G06V40/28—Recognition of hand or arm movements, e.g. recognition of deaf sign language
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/107—Static hand or arm
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Human Computer Interaction (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses the compound convolutional neural networks images of gestures recognition methods under a kind of complex background, this method is to carry out the detection of gesture using the gesture picture training SSD convolutional neural networks under complex background, and criterion exports images of gestures according to area；Then the images of gestures for collecting different gestures carries out specific gesture classification using AlexNet convolutional neural networks model using the images of gestures training AlexNet convolutional neural networks of collection；It will classify from the SSD convolutional neural networks model images of gestures input AlexNet convolutional neural networks model that criterion exports according to area again, export gesture identification result.Carry out detection and localization to the images of gestures under complex background, specific classification may be implemented in this method, realizes the intelligent recognition of images of gestures.
Description
Technical field
Compound convolution mind the present invention relates to Image Acquisition and intelligent identification technology field, under specifically a kind of complex background
Through network images of gestures recognition methods.
Background technique
Gesture is as a kind of direct, efficient man-machine interaction mode, increasingly by everybody attention.Gesture identification application
Become research hotspot instantly, identification of the control of unmanned plane and holder to gesture to unmanned plane, holder image recognition and control
It is required that relatively high, recognition time and discrimination are two key parameters.Traditional Gesture Recognition has Markov model, moves
State time planning, random forest etc..Because traditional identification technology needs Manual definition to extract feature, do not consider that gesture changes yet
Detailed problem in the process, then just not high to the gesture identification discrimination under complex environment, robustness is not also strong.Above-mentioned tradition
Method it is to be improved in the performance indicators such as the identification of hand motion details, recognition time, discrimination.
Summary of the invention
It is an object of the invention to overcome the deficiencies in the prior art, and the compound convolutional Neural under providing a kind of complex background
Carry out detection and localization to the images of gestures under complex background, specific may be implemented in network images of gestures recognition methods, this method
Classification, realizes the intelligent recognition of images of gestures.
Realizing the technical solution of the object of the invention is:
A kind of compound convolutional neural networks images of gestures recognition methods under complex background, includes the following steps:
1) the gesture picture for collecting the images of gestures and different gesture shapes under video acquisition complex background, by the view of acquisition
Frequently each frame interception obtains the gesture picture of the gesture picture and different gesture shapes under complex background at picture；
2) the gesture picture under the complex background obtained to step 1) and the gesture picture of different gesture shapes are located in advance
Reason；
3) the part gesture picture training SSD convolutional neural networks mould under the pretreated complex background of step 2) is utilized
Type, another part gesture picture under the trained pretreated complex background of SSD convolutional neural networks model inspection；
4) gesture for detecting step 3) passes through area decision criteria, and output meets the gesture figure of area decision criteria
Picture；
5) the gesture picture training AlexNet convolutional neural networks of the pretreated different gesture shapes of step 2) are utilized,
Trained AlexNet convolutional neural networks model is exported, the images of gestures of step 4) output is inputted into AlexNet convolutional Neural
Network model classifies to images of gestures using AlexNet convolutional neural networks model, exports gesture identification result.
In step 4), the area decision criteria is the area according to the images of gestures and whole image that detected
Whether ratio-dependent images of gestures can be used as the input images of gestures of disaggregated model；
The images of gestures and the area ratio threshold value of whole image are 1:30, and it is most accurate that area differentiates, i.e., ought detect
Images of gestures and whole image ratio be greater than 1:30, by images of gestures be next gesture classification model AlexNet it is defeated
Enter.
In step 3), the detection is to establish images of gestures data set, and images of gestures is in different tilt angles and not
It is acquired under same light environment, SSD network is sent into after pre-processing to the images of gestures of acquisition, then calculated using gradient decline
Method constantly updates SSD network weight, obtains trained images of gestures detection model.
In step 5), the classification is to establish images of gestures data set, and images of gestures is in different tilt angles and not
It is acquired under same light environment, AlexNet network is sent into after pre-processing to the images of gestures of acquisition, then using under gradient
Algorithm is dropped, AlexNet network weight is constantly updated, obtains trained images of gestures disaggregated model.
The utility model has the advantages that the compound convolutional neural networks images of gestures identification side under a kind of complex background provided by the invention
Method, compared with the conventional method, discrimination is higher, robustness is stronger for this method.
Detailed description of the invention
Fig. 1 is a kind of flow chart of the compound convolutional neural networks images of gestures recognition methods under complex background；
Fig. 2 is the structure chart of SSD convolutional neural networks；
Fig. 3 is the structure chart of AlexNet convolutional neural networks.
Specific embodiment
The present invention is further elaborated with reference to the accompanying drawings and examples, but is not limitation of the invention.
A kind of compound convolutional neural networks images of gestures recognition methods under complex background, includes the following steps:
1) the gesture picture for collecting the images of gestures and different gesture shapes under video acquisition complex background, by the view of acquisition
Frequently each frame interception obtains the gesture picture of the gesture picture and different gesture shapes under complex background at picture；
2) the gesture picture under the complex background obtained to step 1) and the gesture picture of different gesture shapes are located in advance
Reason；
3) the part gesture picture training SSD convolutional neural networks mould under the pretreated complex background of step 2) is utilized
Type, another part gesture picture under the trained pretreated complex background of SSD convolutional neural networks model inspection；
4) gesture for detecting step 3) passes through area decision criteria, and output meets the gesture figure of area decision criteria
Picture；
5) the gesture picture training AlexNet convolutional neural networks of the pretreated different gesture shapes of step 2) are utilized,
Trained AlexNet convolutional neural networks model is exported, the images of gestures of step 4) output is inputted into AlexNet convolutional Neural
Network model classifies to images of gestures using AlexNet convolutional neural networks model, exports gesture identification result.
In step 4), the area decision criteria is the area according to the images of gestures and whole image that detected
Whether ratio-dependent images of gestures can be used as the input images of gestures of disaggregated model；
The images of gestures and the area ratio threshold value of whole image are 1:30, and it is most accurate that area differentiates, i.e., ought detect
Images of gestures and whole image ratio be greater than 1:30, by images of gestures be next gesture classification model AlexNet it is defeated
Enter.
In step 3), the detection is to establish images of gestures data set, and images of gestures is in different tilt angles and not
It is acquired under same light environment, SSD network is sent into after pre-processing to the images of gestures of acquisition, then calculated using gradient decline
Method constantly updates SSD network weight, obtains trained images of gestures detection model.
In step 5), the classification is to establish images of gestures data set, and images of gestures is in different tilt angles and not
It is acquired under same light environment, AlexNet network is sent into after pre-processing to the images of gestures of acquisition, then using under gradient
Algorithm is dropped, AlexNet network weight is constantly updated, obtains trained images of gestures disaggregated model.
The present invention will be described combined with specific embodiments below:
A kind of compound convolutional neural networks images of gestures recognition methods under complex background, is made of software algorithm part,
Software section mainly completes the detection and classification of image, mainly includes the framing of video, the pretreatment of image, the inspection of images of gestures
Survey the calculating of calculating, decision mechanism, the calculating of gesture classification；Gesture of the entire compound convolutional neural networks under complex background is known
Other algorithm flow experimental situation includes server, and CPU takes i7-6700, GPU to take NVIDIA-Titan, SSD convolutional Neural
When network model and AlexNet convolutional neural networks model training, operate under caffe, using Ubuntu16.04 system.Under
Face is further introduced the present invention in conjunction with attached drawing.The structure chart of SSD convolutional neural networks and AlexNet convolutional neural networks
Respectively as shown in Figure 2 and Figure 3, process such as Fig. 1 of the compound convolutional neural networks images of gestures recognition methods under a kind of complex background
It is shown.
In Fig. 2, SSD convolutional neural networks model is made of 4 parts: basic network part, supplementary features extract layer portion
Point, original packet peripheral frame generating portion and convolution predicted portions.Basic network part directly carries out feature extraction to picture, will extract
To feature be directly used in target detection and surround the recurrence (surround frame i.e. of interest object region) of frame, can also will
Feature is input to supplementary features extract layer.VGG-16 model is taken in the basic network part of SSD model, this part includes different big
The picture of input is carried out convolutional calculation, obtains the characteristic pattern that size is unified for 19 × 19 by small convolutional layer.Supplementary features
Extract layer is exactly the increased portion bundling lamination behind facilities network network layers, VGG-16 convolutional layer followed by two full articulamentums,
The target image further feature extracted is input to prediction interval and surrounds frame generation layer.SSD model takes the convolutional layer of part
Characteristic pattern carry out target prediction and surround frame amendment.In order to handle various sizes of image, the same network is taken, is chosen
The characteristic pattern of different convolutional layers surround the recurrence of frame.The characteristic pattern of different convolutional layers has different receptive fields,
The same network, the characteristic value on different characteristic patterns represent various sizes of image block above original image.Specific volume
Lamination is responsible for the object of processing feature size, so needing different zoom factors to correspond to original packet peripheral frame in different convolutional layers
Size, the zoom factor of kth layer is as follows:
Wherein, Smin=0.1, smax=0.8, if the height and width of the original image of input are respectively HinputAnd winput, the layer
Corresponding wide be with height ratioSurround the width and a height of w of framekAnd hk, skFor the ratio for surrounding frame scaling.Image passes through front
Basic network part, supplementary features extract layer and original packet peripheral frame generating portion, the feature being selected are predicted by convolution
Partial operation, obtaining to surround frame correction value and surround inside frame is the probability of target object, to obtain the pre- of target image
Measured value.The true value that images of gestures is obtained by calibration, is then input to loss for the true value of the predicted value of image and image
Function finally carries out the right value update of SSD network using gradient algorithm.
In Fig. 2, gesture classification takes AlexNet model, AlexNet model by volume base, pond layer, full articulamentum and
SoftMax classifier composition.By 227 × 227 picture input volume base, by full articulamentum, the result of prediction and true
Value input loss function, then carries out right value update.
In the network architecture, convolutional layer carries out feature extraction to image by the way that weight is shared, obtains spy by activation primitive
Sign figure.Pond is carried out to obtained characteristic pattern, effect is the generalization ability for enhancing convolutional neural networks.By obtained characteristic pattern into
The full connection of row, obtains feature vector, is input to SoftMax classifier and classifies, that is, assesses the probability value of every one kind, feature to
Amount and be 1, formula is as follows:
Wherein: p (y(i)=j | x(i)；θ) table indicates sample x(i)Belong to the probability of jth class,Indicate the ginseng of model
Number, cost function are as follows:
M is total class number, y in above formula(i)It is the value of jth class, log function is to find maximum probability to sub-category.
As shown in Figure 1, a kind of detailed process of the compound convolutional neural networks images of gestures recognition methods under complex background
Include the following steps: with processing method
1) using the gesture under video camera shooting complex background, the video of each frame is intercepted using opencv, by each frame
Video production at picture, acquire image 10,000 altogether.
2) block diagram marking software is used, the gesture in image is demarcated, the mark of spotting is exported while calibration
Label and the coordinate position in gesture in the picture.
3) picture demarcated is pre-processed, the picture of calibration is cut to 300 × 300, then picture and mark
Label are input to together in forming label file, by documenting at LMDB format.
4) data are input to SSD convolutional neural networks, by calculated predicted value and are really input in Loss function,
The form of Loss function is as follows:
N is the number of matched default boxes, and x is to indicate whether the frame having matched belongs to classification p, and g is true value
Ground truth box, l are prediction blocks, and α is the parameter of prediction；C refers to the confidence level that institute's frame selects target to belong to classification p, root
According to the value of Loss function, using gradient algorithm come the parameter in training convolutional neural networks, following (6) formula institute of gradient algorithm principle
Show:
The vector of difference of the above formula between estimated value and true value indicates that X is to be originally inputted, and θ is model parameter, and y is true
Real value, hθ(x(m)) it is the preceding derivative to input, (x(m))TInput calculating matrix figure, y(m)It is the matrix of the true value of input.Loss
Shown in following (7) formula of the gradient formula of function:
Represent the gradient of loss function, x represents original input value, the parameter of θ representative model, y representative image
True value.
By constantly training iteration to update the weight of SSD convolutional neural networks, until Loss function convergence, SSD mould is exported
Type carries out the detection of images of gestures using SSD.
5) image under complex background condition is inputted into SSD model, detection image is exported by calculating, according to the hand of detection
The coordinate of gesture image cuts out gesture to come.
6) trained SSD model is tested, test accuracy rate 96.7%, thus SSD model there are false retrieval and
Missing inspection, in order to ensure the images of gestures 100% that detected can be used as the input of next gesture classification model, so build again
A gesture area criterion has been found, i.e., gesture has been determined according to the area ratio of the images of gestures and whole image that detected
Whether image can be used as the input images of gestures of disaggregated model；By multiple experiment, images of gestures and whole are schemed in discovery
The proportion threshold value of picture positions 1:30, and it is most accurate that area differentiates, i.e., the ratio of the images of gestures and whole image that ought detect is greater than
1:30, the input for being next gesture classification model AlexNet by images of gestures.
7) gesture classification takes AlexNet model, AlexNet model by volume base, pond layer, full articulamentum and
SoftMax classifier composition, by 227 × 227 picture input volume base, by full articulamentum, the result of prediction and true
Value input loss function, then carries out right value update.
In the network architecture, convolutional layer carries out feature extraction to image by the way that weight is shared, obtains spy by activation primitive
Sign figure carries out pond to obtained characteristic pattern, and effect is to enhance the general Huaneng Group power of convolutional neural networks, the characteristic pattern that will be obtained
It is connected entirely, obtains feature vector, be input to SoftMax classifier and classify.
8) images of gestures that step 6) exports is input in trained AlexNet model, carries out accurate gesture point
Class completes the compound convolutional neural networks images of gestures recognition methods process under entire complex background.
Claims (4)
1. the compound convolutional neural networks images of gestures recognition methods under a kind of complex background, which is characterized in that including walking as follows
It is rapid:
1) the gesture picture of the images of gestures and different gesture shapes under video acquisition complex background is collected, the video of acquisition is every
One frame is intercepted into picture, obtains the gesture picture of the gesture picture and different gesture shapes under complex background；
2) the gesture picture under the complex background obtained to step 1) and the gesture picture of different gesture shapes pre-process；
3) the part gesture picture training SSD convolutional neural networks model under the pretreated complex background of step 2, instruction are utilized
Another part gesture picture under the pretreated complex background of SSD convolutional neural networks model inspection perfected；
4) gesture for detecting step 3) passes through area decision criteria, and output meets the images of gestures of area decision criteria；
5) the gesture picture training AlexNet convolutional neural networks of the pretreated different gesture shapes of step 2, output are utilized
The images of gestures of step 4) output is inputted AlexNet convolutional neural networks by trained AlexNet convolutional neural networks model
Model classifies to images of gestures using AlexNet convolutional neural networks model, exports gesture identification result.
2. the compound convolutional neural networks images of gestures recognition methods under a kind of complex background according to claim 1,
It is characterized in that, in step 4), the area decision criteria is the area according to the images of gestures and whole image that detected
Whether ratio-dependent images of gestures can be used as the input images of gestures of disaggregated model；
The images of gestures and the area ratio threshold value of whole image are 1:30, and area differentiates most accurate, i.e., the hand that ought be detected
The ratio of gesture image and whole image is greater than 1:30, the input for being next gesture classification model AlexNet by images of gestures.
3. the compound convolutional neural networks images of gestures recognition methods under a kind of complex background according to claim 1,
Be characterized in that, in step 3), the detection is to establish images of gestures data set, images of gestures in different tilt angle and
It is acquired under different light environments, SSD network is sent into after pre-processing to the images of gestures of acquisition, then declined using gradient
Algorithm constantly updates SSD network weight, obtains trained images of gestures detection model.
4. the compound convolutional neural networks images of gestures recognition methods under a kind of complex background according to claim 1,
Be characterized in that, in step 5), the classification is to establish images of gestures data set, images of gestures in different tilt angle and
It is acquired under different light environments, AlexNet network is sent into after pre-processing to the images of gestures of acquisition, then utilize gradient
Descent algorithm constantly updates AlexNet network weight, obtains trained images of gestures disaggregated model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811534388.8A CN109635750A (en) | 2018-12-14 | 2018-12-14 | A kind of compound convolutional neural networks images of gestures recognition methods under complex background |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811534388.8A CN109635750A (en) | 2018-12-14 | 2018-12-14 | A kind of compound convolutional neural networks images of gestures recognition methods under complex background |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109635750A true CN109635750A (en) | 2019-04-16 |
Family
ID=66074186
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811534388.8A Pending CN109635750A (en) | 2018-12-14 | 2018-12-14 | A kind of compound convolutional neural networks images of gestures recognition methods under complex background |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109635750A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110110665A (en) * | 2019-05-08 | 2019-08-09 | 北京创源微致软件有限公司 | The detection method of hand region under a kind of driving environment |
| CN110414402A (en) * | 2019-07-22 | 2019-11-05 | 北京达佳互联信息技术有限公司 | A kind of gesture data mask method, device, electronic equipment and storage medium |
| CN110674741A (en) * | 2019-09-24 | 2020-01-10 | 广西师范大学 | Machine vision gesture recognition method based on dual-channel feature fusion |
| CN110968199A (en) * | 2019-12-10 | 2020-04-07 | 安徽师范大学 | Traffic light control method and system based on gesture recognition and electronic equipment |
| CN111523435A (en) * | 2020-04-20 | 2020-08-11 | 安徽中科首脑智能医疗研究院有限公司 | Finger detection method, system and storage medium based on target detection SSD |
| CN111736702A (en) * | 2019-06-27 | 2020-10-02 | 谷歌有限责任公司 | Intent detection with computing devices |
| CN112396666A (en) * | 2020-12-09 | 2021-02-23 | 广西双英集团股份有限公司 | Intelligent assembling process control method based on gesture recognition |
| CN113033290A (en) * | 2021-02-01 | 2021-06-25 | 广州朗国电子科技有限公司 | Image subregion identification method, device and storage medium |
| CN113703581A (en) * | 2021-09-03 | 2021-11-26 | 广州朗国电子科技股份有限公司 | Window adjusting method based on gesture switching, electronic whiteboard and storage medium |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101853071A (en) * | 2010-05-13 | 2010-10-06 | 重庆大学 | Gesture identification method and system based on visual sense |
| WO2016026063A1 (en) * | 2014-08-21 | 2016-02-25 | Xiaoou Tang | A method and a system for facial landmark detection based on multi-task |
| CN107657233A (en) * | 2017-09-28 | 2018-02-02 | 东华大学 | Static sign language real-time identification method based on modified single multi-target detection device |
| CN108073851A (en) * | 2016-11-08 | 2018-05-25 | 株式会社理光 | A kind of method, apparatus and electronic equipment for capturing gesture identification |
| CN108460329A (en) * | 2018-01-15 | 2018-08-28 | 任俊芬 | A kind of face gesture cooperation verification method based on deep learning detection |
| CN108549852A (en) * | 2018-03-28 | 2018-09-18 | 中山大学 | Pedestrian detector's Auto-learning Method under special scenes based on the enhancing of depth network |
| CN108875911A (en) * | 2018-05-25 | 2018-11-23 | 同济大学 | One kind is parked position detecting method |
| CN108932500A (en) * | 2018-07-09 | 2018-12-04 | 广州智能装备研究院有限公司 | A kind of dynamic gesture identification method and system based on deep neural network |
-
2018
- 2018-12-14 CN CN201811534388.8A patent/CN109635750A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101853071A (en) * | 2010-05-13 | 2010-10-06 | 重庆大学 | Gesture identification method and system based on visual sense |
| WO2016026063A1 (en) * | 2014-08-21 | 2016-02-25 | Xiaoou Tang | A method and a system for facial landmark detection based on multi-task |
| CN108073851A (en) * | 2016-11-08 | 2018-05-25 | 株式会社理光 | A kind of method, apparatus and electronic equipment for capturing gesture identification |
| CN107657233A (en) * | 2017-09-28 | 2018-02-02 | 东华大学 | Static sign language real-time identification method based on modified single multi-target detection device |
| CN108460329A (en) * | 2018-01-15 | 2018-08-28 | 任俊芬 | A kind of face gesture cooperation verification method based on deep learning detection |
| CN108549852A (en) * | 2018-03-28 | 2018-09-18 | 中山大学 | Pedestrian detector's Auto-learning Method under special scenes based on the enhancing of depth network |
| CN108875911A (en) * | 2018-05-25 | 2018-11-23 | 同济大学 | One kind is parked position detecting method |
| CN108932500A (en) * | 2018-07-09 | 2018-12-04 | 广州智能装备研究院有限公司 | A kind of dynamic gesture identification method and system based on deep neural network |
Non-Patent Citations (3)
| Title |
|---|
| XIONG XU 等: "A new deep learning model based on improved alexnet for radiation source target recognition", 《ACM》 * |
| 张勋 等: "一种基于深度学习的静态手势实时识别方法", 《现代计算机》 * |
| 苏思悦 等: "结合单目标多窗口检测器和AlexNet的螺母检测", 《中国科技论文》 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110110665A (en) * | 2019-05-08 | 2019-08-09 | 北京创源微致软件有限公司 | The detection method of hand region under a kind of driving environment |
| CN110110665B (en) * | 2019-05-08 | 2021-05-04 | 北京创源微致软件有限公司 | Detection method for hand area in driving environment |
| CN111736702A (en) * | 2019-06-27 | 2020-10-02 | 谷歌有限责任公司 | Intent detection with computing devices |
| US11960793B2 (en) | 2019-06-27 | 2024-04-16 | Google Llc | Intent detection with a computing device |
| CN110414402B (en) * | 2019-07-22 | 2022-03-25 | 北京达佳互联信息技术有限公司 | Gesture data labeling method and device, electronic equipment and storage medium |
| CN110414402A (en) * | 2019-07-22 | 2019-11-05 | 北京达佳互联信息技术有限公司 | A kind of gesture data mask method, device, electronic equipment and storage medium |
| CN110674741A (en) * | 2019-09-24 | 2020-01-10 | 广西师范大学 | Machine vision gesture recognition method based on dual-channel feature fusion |
| CN110674741B (en) * | 2019-09-24 | 2023-06-09 | 青岛海之晨工业装备有限公司 | Gesture recognition method in machine vision based on double-channel feature fusion |
| CN110968199A (en) * | 2019-12-10 | 2020-04-07 | 安徽师范大学 | Traffic light control method and system based on gesture recognition and electronic equipment |
| CN111523435A (en) * | 2020-04-20 | 2020-08-11 | 安徽中科首脑智能医疗研究院有限公司 | Finger detection method, system and storage medium based on target detection SSD |
| CN112396666A (en) * | 2020-12-09 | 2021-02-23 | 广西双英集团股份有限公司 | Intelligent assembling process control method based on gesture recognition |
| CN113033290A (en) * | 2021-02-01 | 2021-06-25 | 广州朗国电子科技有限公司 | Image subregion identification method, device and storage medium |
| CN113703581A (en) * | 2021-09-03 | 2021-11-26 | 广州朗国电子科技股份有限公司 | Window adjusting method based on gesture switching, electronic whiteboard and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109635750A (en) | A kind of compound convolutional neural networks images of gestures recognition methods under complex background | |
| Long et al. | Safety helmet wearing detection based on deep learning | |
| Zhang et al. | Deep convolutional neural networks for forest fire detection | |
| CN108961235A (en) | A kind of disordered insulator recognition methods based on YOLOv3 network and particle filter algorithm | |
| CN105718912B (en) | A kind of vehicle characteristics object detecting method based on deep learning | |
| Kang et al. | Voxel-based extraction and classification of 3-D pole-like objects from mobile LiDAR point cloud data | |
| CN107688830B (en) | Generation method of vision information correlation layer for case serial-parallel | |
| CN110826424A (en) | Pedestrian searching method based on pedestrian re-identification driving positioning adjustment | |
| CN106503170B (en) | It is a kind of based on the image base construction method for blocking dimension | |
| CN109344842A (en) | A kind of pedestrian's recognition methods again based on semantic region expression | |
| CN111145222A (en) | Fire detection method combining smoke movement trend and textural features | |
| CN113657423A (en) | Target detection method suitable for small-volume parts and stacked parts and application thereof | |
| CN115719445A (en) | Seafood identification method based on deep learning and raspberry type 4B module | |
| CN116109950A (en) | Low-airspace anti-unmanned aerial vehicle visual detection, identification and tracking method | |
| Wang et al. | Forest smoke detection based on deep learning and background modeling | |
| CN111354016A (en) | Unmanned aerial vehicle ship tracking method and system based on deep learning and difference value hashing | |
| CN111144209A (en) | Monitoring video head detection method based on heterogeneous multi-branch deep convolutional neural network | |
| CN111767826A (en) | Timing fixed-point scene abnormity detection method | |
| CN116129327A (en) | Infrared vehicle detection method based on improved YOLOv7 algorithm | |
| CN109492513B (en) | Face space duplication eliminating method for light field monitoring | |
| Wu et al. | Deep-rice: Deep multi-sensor image recognition for grading rice | |
| Liu et al. | Pedestrian detection method in infrared image based on improved YOLOv7 | |
| CN112069997B (en) | Unmanned aerial vehicle autonomous landing target extraction method and device based on DenseHR-Net | |
| Zhang et al. | Safety Helmet and Mask Detection at Construction Site Based on Deep Learning | |
| CN113496210A (en) | Attention mechanism-based photovoltaic string tracking and fault tracking method |
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: 20190416 |