CN1818930A - Eyes open detection with multi-nerve network combination based on identifying model - Google Patents
Eyes open detection with multi-nerve network combination based on identifying model Download PDFInfo
- Publication number
- CN1818930A CN1818930A CN 200610024977 CN200610024977A CN1818930A CN 1818930 A CN1818930 A CN 1818930A CN 200610024977 CN200610024977 CN 200610024977 CN 200610024977 A CN200610024977 A CN 200610024977A CN 1818930 A CN1818930 A CN 1818930A
- Authority
- CN
- China
- Prior art keywords
- human eye
- neural network
- model
- eyes
- cognition
- 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.)
- Granted
Links
Images
Abstract
A multi-nerve network combinative opening-eyes examination method base on distinguishing model belongs to the field of image process. The invention includes: A, marking the connecting area after binary, filtrate the possible area according to the binary geometry module of human eyes; B, designing the multi-nerve network combinative detector base on the distinguishing module, namely distinguish the binary image by the radial basic nerve network, detect the opening-eyes if it distinguish, otherwise carry through the next step. C, distinguishing the ash-image by the anti-nerve network, cumulating the times if it undistinguished, that will not detect the opening-eyes if the times are over six, otherwise continuing the next step. D, detecting the opening-eyes if the studying state is not set, asking the area whether or not is human eyes to the teacher if the studying state is set, saving the binary image and retraining the radial basic nerve network if it is human eyes, otherwise saving the ash-image and retraining the anti-nerve network.The invention does not need lots of original training samples, the detecting capability will improve continuously by the supervised study.
Description
Technical field
What the present invention relates to is a kind of method of technical field of image processing, and specifically, what relate to is a kind of detection method of opening eyes of the multiple neural network combination based on model of cognition.
Background technology
Whether the problem that the method that the present invention proposes solves is after obtaining facial image, how to discern its eyes and open.The identification people state of whether opening eyes can be widely used in all kinds of intelligent interactive systems, as in the network remote teaching to the detection of attention of student, in the car steering to the detection of driver's notice.Yet all multifactor influences such as illumination, size, attitude, glasses, eyebrow make this identification work become one and have challenging task.The work relevant with eyes at present mainly concentrates on the location to eyes, promptly find out and open or the position of closed eyes, main method is to utilize the priori of eyes, as colouring information, shape information, distributed intelligence etc., carry out feature extraction, after obtaining feature with training sample,, finish location work by matching characteristic and then identification eyes.
Find through literature search prior art, article Eye detection using color cues andprojection functions in Proc.2002Int.Conf.on Image Processing, 2002, vol.3 (adopts the eye detection method of colouring information and projection function, image processing international conference in 2002) proposes to utilize colouring information to find out skin area earlier, again skin area and near the searching eyes, but this method is subjected to the influence of illumination easily.Article Detecting and tracking eye by using theirphysiological properties, dynamics and appearance in Proc.Of IEEE Conf.on Computer Vision and Pattern Recognition, 2000, vol.1 (utilizes the eye detection and the tracking of physiological property, computer vision and pattern-recognition IEEE meeting in 2000) a kind of eye detection method that adopts infrared illumination has been described, but system depends on the restriction of other object infrared external reflections to external world.Article Robust eye extraction using deformable template and featuretracking ability in Proc.of the Joint Conf.of the Fourth Int.Conf.on Information, Communications and Signal Processing, and the FourthPacific Rim Conf.on Multimedia, 2003, vol.3 (adopts the eyes dividing method of variable template and signature tracking, the 4th information, communication and signal Processing and Pacific Rim multimedia joint conference) template matches is applied to eye detection, yet template matches is not only calculated length consuming time, and closely related with the selection of eyes template.Because the diversity and the polytrope of real world are even a large amount of training samples also can't be contained all situations.
Summary of the invention
The present invention is directed to the deficiencies in the prior art, on the basis of using for reference people's identifying, a kind of model of cognition has been proposed, and on this model basis, provide a kind of detection method of opening eyes of the multiple neural network combination based on model of cognition, the study of supervision is arranged, and initial training sample that need not be a large amount of, can realize the real-time detection of opening eyes, and, detect performance and can obtain continuing to improve along with study constantly.
The present invention is achieved through the following technical solutions, and may further comprise the steps:
A, people's face coloured image after binary conversion treatment, carry out mark to connected region, and under the guidance of human eye binary map geometric model, the zone that screening is communicated with obtains possible human eye area.
B, to possible human eye area, carry out normalization, the detecting device that designs following multiple neural network combination based on model of cognition detects again.Promptly, the binary map after its normalization is discerned with the radial base neural net of training (RBF), then thought to detect and open eyes if recognize human eye to possible human eye area.If do not recognize human eye, then carry out next step.
C, to possible human eye area, gray-scale map after its normalization is discerned with the reverse neural network of training (BP),, then added up not detect continuously the number of times of opening eyes if do not recognize human eye, if do not detect the number of times of opening eyes greater than 6, then think not detect and open eyes.If recognize human eye, then carry out next step.
D, if be not set to the supervised learning state, then think to detect and open eyes.If be set to the supervised learning state, then mutual with the teacher, inquire whether this zone is human eye.If answering is then to preserve this human eye binary map, train RBF Neural Network again.If answer not, then preserve this human eye gray-scale map, train reverse neural network again.
Described under the guidance of human eye binary map geometric model, the zone that screening is communicated with, obtain possible human eye area, be meant: four quasi-modes that utilize human eye binary map geometric model: up and down, about, tiltedly to independent pattern, all connected regions are screened, stay the connected region that meets one of four quasi-modes.
Described model of cognition is promptly imported object to be identified, at first according to the memory knowledge of accumulating, judge and whether to meet, if judging to meet then thinks identification, as if not meeting, then, judge whether identification, can not discern if judge again according to the inferenctial knowledge of accumulation, then think and to discern,, then seek advice from if judge identification, after the answer that obtains being, then think identification, and upgrade memory knowledge, after obtaining answer not, then think and to discern, and upgrade inferenctial knowledge.
Described multiple neural network combination, be meant: to possible human eye area, binary map after its normalization is discerned with the radial base neural net of training, if output and { 0, the Euclidean distance of 1} is greater than 0.06, then after the gray-scale map normalization with the possibility human eye area, discern with the reverse neural network of training.
Described radial base neural net is meant: be made up of radially basic neuron layer and output layer, input vector again by the BP neuron computes, gets output to the end by after the radially basic neuron computes.
Described binary map after its normalization is discerned with the radial base neural net of training, be meant: the binary map to possible human eye area is done normalized.Binary map after the normalization as input, is judged the human eye whether open in this zone by the output of radial base neural net.
Described gray-scale map after its normalization is discerned with the reverse neural network of training, be meant: the gray-scale map to possible human eye area is done normalized.Gray-scale map after the normalization as input, is judged the human eye whether open in this zone by the output of reverse neural network.
Described supervised learning, promptly in identifying, mutual with the teacher, inquire whether this zone is human eye, if answer be, then preserve this human eye binary map, automatically train RBF Neural Network again, if answer not, then preserve this human eye gray-scale map, train reverse neural network automatically again.
The present invention is on the model of cognition basis that is proposed, proposition is in conjunction with the detection method of opening eyes of multiple neural network, " memory " function of radial base neural net and " reasoning " function of reverse neural network have been brought into play, initial training sample that need not be a large amount of, and, detect performance and can obtain continuing to improve along with supervised learning constantly.
Description of drawings
Fig. 1 is a building-block of logic of the present invention.
Fig. 2 is the synoptic diagram of human eye binary map geometric model among the present invention.
Wherein, (a) be pattern up and down; (b) be left and right sides pattern; (c-1), (c-2) is for tiltedly to pattern; (d-1), (d-2) is independent pattern.
Fig. 3 is the synoptic diagram of model of cognition among the present invention.
Fig. 4 is the synoptic diagram of radial base neural net among the present invention.
Fig. 5 is the synoptic diagram of reverse neural network among the present invention.
Fig. 6 is the synoptic diagram of neural network training part initial training sample among the present invention.
Wherein, (a-1), (a-2) is the part initial training sample of radial base neural net; (b-1), (b-2) is the part initial training sample of reverse neural network.
Fig. 7 is the exemplary plot of some testing results of opening eyes of the embodiment of the invention.
Embodiment
Provide embodiment below in conjunction with accompanying drawing and technical solution of the present invention:
As shown in Figure 1, people's face coloured image at first carries out binary conversion treatment, promptly by maximum variance between clusters (OSTU) gradation of image value scope is changed into 0 and 1.Connected region is carried out mark, be about in the image each other about, up and down or oblique adjacent gray-scale value be 1 pixel, be labeled as same connected region, thereby obtain one group of alternative area.Under the guidance of human eye binary map geometric model such as Fig. 2, the screening alternative area stays the connected region that meets one of four quasi-modes, obtains possible human eye area.To possible human eye area, according to model of cognition such as Fig. 3, the detecting device that designs following multiple neural network combination detects again.Promptly, the binary map after its normalization with radial base neural net such as Fig. 4 of training, is discerned possible human eye area.Normalization promptly adopts linear interpolation method that image is changed to wide 36 pixels, the fixed size of high 20 pixels dimensionally.If radial base neural net output is with { 0, the Euclidean distance of 1} thinks then that less than 0.06 this zone is the human eye of opening.If output is with { 0, the Euclidean distance of 1} is greater than 0.06, and then after the gray-scale map normalization with the possibility human eye area, reverse neural network such as Fig. 5 with training discern.The gray-scale map gray values of pixel points is calculated by the coloured image corresponding pixel points: and 0.299* is red+and 0.587* is green+0.114 indigo plant.If reverse neural network output is with { 0, the Euclidean distance of 1} then adds up not detect the number of times of opening eyes continuously greater than 0.01, if do not detect the number of times of opening eyes greater than 6, then thinks not detect and opens eyes.If reverse neural network output is with { 0, the Euclidean distance of 1} is less than 0.01, and is not provided with the supervised learning state, thinks that then this zone is the human eye of opening.If reverse neural network output is with { 0, the Euclidean distance of 1} is less than 0.01, and is provided with the supervised learning state, then inquires the teacher, and whether this possible human eye area opens eyes.If answer be, binary map that then should the zone saves as learning sample, automatically train RBF Neural Network again.If answer not, gray-scale map that then should the zone saves as learning sample, trains reverse neural network automatically again.
As shown in Figure 2, four quasi-modes of human eye binary map geometric model: up and down, about, tiltedly to independent pattern.Pattern up and down, promptly the horizontal range at two connected region centers less than its horizontal width and 1/2nd, simultaneously vertical range greater than its vertical height and 1/2nd, and less than its vertical height and 3/2nds.Left and right sides pattern, promptly the vertical range at two connected region centers less than its vertical height and, simultaneously horizontal range greater than its horizontal width and 1/2nd, and less than its horizontal width and four times.Tiltedly to pattern, promptly the horizontal range at two connected region centers greater than its horizontal width and 1/2nd, and less than its horizontal width and three times, simultaneously vertical range greater than its vertical height and 1/2nd, and less than its vertical height and.Independent pattern, promptly the horizontal range apart from certain connected region center is the position of 1.4 times of its horizontal widths, exists with this connected region similarly regional.With the similar zone of this connected region, promptly gray-scale value is that to account for all gray-scale values in this connected region be that the ratio of 1 pixel is greater than 90% for 1 corresponding pixel points in the zone.
As shown in Figure 3, import object to be identified, at first, judge and whether met, then think and to discern if judge to meet according to the memory knowledge of accumulation.If do not meet, then, judge whether and can discern again according to the inferenctial knowledge of accumulation, if judge and can not discern, then think and can't discern.If judge and can discern, then seek advice from, after the answer that obtains being, then think and can discern, and upgrade memory knowledge.After obtaining answer not, then think and to discern, and upgrade inferenctial knowledge.
As shown in Figure 4, radial base neural net is made up of radially basic neuron layer and output layer.Input vector again by the BP neuron computes, gets output to the end by after the radially basic neuron computes.The output of radial base neural net is designed to: the representative do not detect open eyes 1,0} and the representative detect open eyes 0,1}.Learning parameter is 0.8, and the study threshold value is 0.06.
As shown in Figure 5, reverse neural network is made up of hidden layer and output layer.Input vector by the BP neuron computes of output layer, gets output to the end again by after the BP neuron computes of hidden layer.The output of reverse neural network is designed to: the representative do not detect open eyes 1,0} and the representative detect open eyes 0,1}.The hidden neuron number is 60, and learning parameter is 0.8, and the study threshold value is 0.01.
As shown in Figure 6, the part initial training sample of train RBF Neural Network and reverse neural network.(a-1) be the initial human eye training sample of radial base neural net part; (a-2) be the initial non-human eye training sample of radial base neural net part.(b-1) be the initial human eye training sample of reverse neural network part; (b-2) be the initial non-human eye training sample of reverse neural network part.
As shown in Figure 7, some of embodiment of the invention testing result examples of opening eyes.Detect open eyes after, with red rectangle frame represent the to open eyes zone at place.
As seen from the above embodiment: the initial training sample that the present invention need not be a large amount of, can realize the real-time detection of opening eyes, and, detect performance and can obtain continuing to improve along with study constantly.
Claims (9)
1. the detection method of opening eyes based on the multiple neural network combination of model of cognition is characterized in that, may further comprise the steps:
A, people's face coloured image after binary conversion treatment, carry out mark to connected region, and under the guidance of human eye binary map geometric model, the zone that screening is communicated with obtains possible human eye area;
B, to possible human eye area, detecting device based on the combination of model of cognition design multiple neural network detects, promptly to possible human eye area, binary map after its normalization is discerned with the radial base neural net of training, if recognizing human eye then thinks to detect and opens eyes, if do not recognize human eye, then carry out next step;
C, to possible human eye area, gray-scale map after its normalization is discerned with the reverse neural network of training, if do not recognize human eye, then add up not detect continuously the number of times of opening eyes, if do not detect the number of times of opening eyes greater than 6, then think not detect and open eyes,, then carry out next step if recognize human eye;
D, if be not set to the supervised learning state, then think to detect and open eyes, then mutual if be set to the supervised learning state with the teacher, carry out supervised learning.
2. the eye detection method of the multiple neural network combination based on model of cognition according to claim 1, it is characterized in that, described under the guidance of human eye binary map geometric model, the zone that screening is communicated with, obtain possible human eye area, be meant: four quasi-modes that utilize human eye binary map geometric model: up and down, about, tiltedly to independent pattern, all connected regions are screened, stay the connected region that meets one of four quasi-modes.
3. the eye detection method of the multiple neural network combination based on model of cognition according to claim 2 is characterized in that four quasi-modes of described human eye binary map geometric model: up and down, about, tiltedly to independent pattern, be specially:
Pattern up and down, promptly the horizontal range at two connected region centers less than its horizontal width and 1/2nd, simultaneously vertical range greater than its vertical height and 1/2nd, and less than its vertical height and 3/2nds;
Left and right sides pattern, promptly the vertical range at two connected region centers less than its vertical height and, simultaneously horizontal range greater than its horizontal width and 1/2nd, and less than its horizontal width and four times;
Tiltedly to pattern, promptly the horizontal range at two connected region centers greater than its horizontal width and 1/2nd, and less than its horizontal width and three times, simultaneously vertical range greater than its vertical height and 1/2nd, and less than its vertical height and;
Independent pattern, promptly the horizontal range apart from certain connected region center is the position of 1.4 times of its horizontal widths, exists with this connected region similarly regional;
With the similar zone of this connected region, promptly gray-scale value is that to account for all gray-scale values in this connected region be that the ratio of 1 pixel is greater than 90% for 1 corresponding pixel points in the zone.
4. the eye detection method of the multiple neural network combination based on model of cognition according to claim 1, it is characterized in that, described binary map after its normalization is discerned with the radial base neural net of training, be meant: the binary map to possible human eye area is done normalized, binary map after the normalization as input, is judged the human eye whether open in this zone by the output of radial base neural net.
5. the eye detection method of the multiple neural network combination based on model of cognition according to claim 1, it is characterized in that, described gray-scale map after its normalization is discerned with the reverse neural network of training, be meant: the gray-scale map to possible human eye area is done normalized, gray-scale map after the normalization as input, is judged the human eye whether open in this zone by the output of reverse neural network.
6. the eye detection method of the multiple neural network combination based on model of cognition according to claim 1, it is characterized in that described model of cognition is promptly imported object to be identified, at first according to the memory knowledge of accumulating, judge and whether to meet, if judging to meet then thinks identification, as if not meeting, then again according to the inferenctial knowledge of accumulating, judge whether identification,, then think and to discern if judge and to discern, if judge identification, then seek advice from, after the answer that obtains being, then think identification, and upgrade and remember knowledge, after obtaining answer not, then think and to discern, and upgrade inferenctial knowledge.
7. the eye detection method of the multiple neural network combination based on model of cognition according to claim 1, it is characterized in that, described multiple neural network combination, be meant: to possible human eye area, binary map after its normalization is discerned with the radial base neural net of training, if output is with { 0, the Euclidean distance of 1} is greater than 0.06, then after the gray-scale map normalization with the possibility human eye area, discern with the reverse neural network of training.
8. according to the eye detection method of claim 1 or 7 described multiple neural network combinations based on model of cognition, it is characterized in that, described radial base neural net, be meant: form by radially basic neuron layer and output layer, input vector is by after the radially basic neuron computes, by the BP neuron computes, get output to the end again.
9. the eye detection method of the multiple neural network combination based on model of cognition according to claim 1, it is characterized in that described supervised learning is promptly in identifying, mutual with the teacher, inquiring whether this zone is human eye, is then to preserve this human eye binary map if answer, automatically train RBF Neural Network again, if answer not, then preserve this human eye gray-scale map, train reverse neural network automatically again.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2006100249772A CN100373400C (en) | 2006-03-23 | 2006-03-23 | Eyes open detection with multi-nerve network combination based on identifying model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2006100249772A CN100373400C (en) | 2006-03-23 | 2006-03-23 | Eyes open detection with multi-nerve network combination based on identifying model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN1818930A true CN1818930A (en) | 2006-08-16 |
| CN100373400C CN100373400C (en) | 2008-03-05 |
Family
ID=36918937
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNB2006100249772A Expired - Fee Related CN100373400C (en) | 2006-03-23 | 2006-03-23 | Eyes open detection with multi-nerve network combination based on identifying model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN100373400C (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN100463000C (en) * | 2007-09-27 | 2009-02-18 | 上海交通大学 | Human eye state detection method based on cascade classification and hough circle transform |
| CN102306306A (en) * | 2011-08-30 | 2012-01-04 | 西安交通大学 | Character recognition method for carrying out super-resolution reconstruction by utilizing characteristics of local binary pattern and radial basis function |
| CN102339384A (en) * | 2011-09-16 | 2012-02-01 | 北京交通大学 | Face recognition method |
| CN101520842B (en) * | 2008-02-29 | 2013-05-08 | 佳能株式会社 | Information processing apparatus, eye open/closed degree determination method and image sensing apparatus |
| CN108960071A (en) * | 2018-06-06 | 2018-12-07 | 武汉幻视智能科技有限公司 | A kind of eye opening closed-eye state detection method |
| CN110221699A (en) * | 2019-06-13 | 2019-09-10 | 北京师范大学珠海分校 | A kind of eye movement Activity recognition method of front camera video source |
| CN110334697A (en) * | 2018-08-11 | 2019-10-15 | 昆山美卓智能科技有限公司 | Intelligent table, monitoring system server and monitoring method with condition monitoring function |
| CN111753793A (en) * | 2020-06-30 | 2020-10-09 | 重庆紫光华山智安科技有限公司 | Model training method and device, face screening method and electronic equipment |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP3769442B2 (en) * | 2000-02-15 | 2006-04-26 | ナイルス株式会社 | Eye state detection device |
| JP2001307076A (en) * | 2000-04-24 | 2001-11-02 | Niles Parts Co Ltd | Eye state detector and nap detector |
| JP3852294B2 (en) * | 2001-03-22 | 2006-11-29 | 日産自動車株式会社 | Dozing detection device |
| JP3855790B2 (en) * | 2002-02-12 | 2006-12-13 | 日産自動車株式会社 | Driver status detection device |
| CN1225375C (en) * | 2003-07-02 | 2005-11-02 | 北京交通大学 | Method for detecting fatigue driving based on multiple characteristic fusion |
-
2006
- 2006-03-23 CN CNB2006100249772A patent/CN100373400C/en not_active Expired - Fee Related
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN100463000C (en) * | 2007-09-27 | 2009-02-18 | 上海交通大学 | Human eye state detection method based on cascade classification and hough circle transform |
| CN101520842B (en) * | 2008-02-29 | 2013-05-08 | 佳能株式会社 | Information processing apparatus, eye open/closed degree determination method and image sensing apparatus |
| CN102306306A (en) * | 2011-08-30 | 2012-01-04 | 西安交通大学 | Character recognition method for carrying out super-resolution reconstruction by utilizing characteristics of local binary pattern and radial basis function |
| CN102306306B (en) * | 2011-08-30 | 2013-01-02 | 西安交通大学 | Character recognition method for carrying out super-resolution reconstruction by utilizing characteristics of local binary pattern and radial basis function |
| CN102339384A (en) * | 2011-09-16 | 2012-02-01 | 北京交通大学 | Face recognition method |
| CN102339384B (en) * | 2011-09-16 | 2013-07-03 | 北京交通大学 | Face recognition method |
| CN108960071A (en) * | 2018-06-06 | 2018-12-07 | 武汉幻视智能科技有限公司 | A kind of eye opening closed-eye state detection method |
| CN110334697A (en) * | 2018-08-11 | 2019-10-15 | 昆山美卓智能科技有限公司 | Intelligent table, monitoring system server and monitoring method with condition monitoring function |
| CN110221699A (en) * | 2019-06-13 | 2019-09-10 | 北京师范大学珠海分校 | A kind of eye movement Activity recognition method of front camera video source |
| CN110221699B (en) * | 2019-06-13 | 2022-03-25 | 北京师范大学珠海分校 | Eye movement behavior identification method of front-facing camera video source |
| CN111753793A (en) * | 2020-06-30 | 2020-10-09 | 重庆紫光华山智安科技有限公司 | Model training method and device, face screening method and electronic equipment |
| CN111753793B (en) * | 2020-06-30 | 2022-11-22 | 重庆紫光华山智安科技有限公司 | Model training method and device, face screening method and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN100373400C (en) | 2008-03-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN1818930A (en) | Eyes open detection with multi-nerve network combination based on identifying model | |
| Sivaraman et al. | A general active-learning framework for on-road vehicle recognition and tracking | |
| CN103473950B (en) | Parking lot parking space monitoring method | |
| CN104504395A (en) | Method and system for achieving classification of pedestrians and vehicles based on neural network | |
| CN104268583B (en) | Pedestrian re-recognition method and system based on color area features | |
| CN104517103A (en) | Traffic sign classification method based on deep neural network | |
| CN110119726B (en) | Vehicle brand multi-angle identification method based on YOLOv3 model | |
| CN101604382A (en) | A kind of learning fatigue recognition interference method based on human facial expression recognition | |
| CN106650786A (en) | Image recognition method based on multi-column convolutional neural network fuzzy evaluation | |
| CN106096602A (en) | A kind of Chinese licence plate recognition method based on convolutional neural networks | |
| CN112395951B (en) | Complex scene-oriented domain-adaptive traffic target detection and identification method | |
| CN106897673A (en) | A kind of recognition methods again of the pedestrian based on retinex algorithms and convolutional neural networks | |
| CN111563452A (en) | Multi-human body posture detection and state discrimination method based on example segmentation | |
| CN110032932B (en) | Human body posture identification method based on video processing and decision tree set threshold | |
| CN111860274A (en) | Traffic police command gesture recognition method based on head orientation and upper half body skeleton characteristics | |
| CN105138974A (en) | Gabor coding based finger multimodal feature fusion method | |
| CN111401188A (en) | Traffic police gesture recognition method based on human body key point features | |
| CN114399838A (en) | Multi-person behavior recognition method and system based on attitude estimation and double classification | |
| CN113723277B (en) | Learning intention monitoring method and system integrated with multi-mode visual information | |
| CN116052222A (en) | Cattle face recognition method for naturally collecting cattle face image | |
| Barodi et al. | An enhanced artificial intelligence-based approach applied to vehicular traffic signs detection and road safety enhancement | |
| CN105893941A (en) | Facial expression identifying method based on regional images | |
| Long | A Lightweight Face Recognition Model Using Convolutional Neural Network for Monitoring Students in E-Learning. | |
| CN112329541B (en) | Crowd behavior identification method based on storyboard relation model | |
| CN109284752A (en) | A kind of rapid detection method of vehicle |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| C17 | Cessation of patent right | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20080305 Termination date: 20130323 |