CN106127123A - A kind of human pilot face real-time detection method of driving round the clock based on RGB I - Google Patents
A kind of human pilot face real-time detection method of driving round the clock based on RGB I Download PDFInfo
- Publication number
- CN106127123A CN106127123A CN201610436625.1A CN201610436625A CN106127123A CN 106127123 A CN106127123 A CN 106127123A CN 201610436625 A CN201610436625 A CN 201610436625A CN 106127123 A CN106127123 A CN 106127123A
- Authority
- CN
- China
- Prior art keywords
- convnet
- face
- model
- driver
- rgb
- 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
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
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
-
- 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
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/172—Classification, e.g. identification
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of human pilot face real-time detection method of driving round the clock based on RGB I, including step and the step of driver's face recognition of model training；The step of model training includes: S1, the pretreatment of driver's face database: driver's face picture is divided into two groups of sizes and is respectively 40*40 and 28*28, carry out gray processing afterwards；S2, set up R Convnet model based on RGB image training and I Convnet model based on the training of Infrared image, R Convnet model is divided into R40 Convnet model and R28 Convnet model, I Convnet model is divided into I40 Convnet model, I28 Convnet model, 4 kinds of Convnet models to be to remove full articulamentum on the basis of CNN to obtain；The step of driver's face recognition includes: S3, utilizes RGB photographic head, Infrared photographic head to gather driver's image the most by day with night；S4, is identified with the R40 Convnet model of the cascade trained, R28 Convnet model daytime, uses night the I40 Convnet model of cascade trained, I28 Convnet model to be identified.The present invention improves under night or severe driving environment, the time of driver's face detection and precision.
Description
Technical field
A kind of method that the present invention relates to real-time detection driver face, refers in particular to a kind of driving round the clock based on RGB-I and drives
Sail personnel's face real-time detection method.
Background technology
In recent years, people's living standard day by day improves, and the quantity having private car is continuously increased, sending out of vehicle accident simultaneously
Raw rate is also constantly rising, and vehicle and safety problem have become as the focus of social concerns.Safety problem includes driver fatigue
Detect, drive when intoxicated, the detection of driver's emotion, driving behavior detection etc., and in driver fatigue detection and expression detection
First have to carry out the detection (Face datection) of face location.
Face datection refers to determine the position of all face (if present), size, the process of attitude in the input image.
Face datection problem is originally derived from Face datection, and the research of Face datection can trace back to the 60-70 age in 20th century, Jing Guoji
The tortuous development of 10 years is the most ripe.Face datection is a key link in automatic face detection system, but in early days
Face datection research will be for having the facial image (image as without background) of relatively Condition of Strong Constraint, face location is easy to
Obtain, so Face datection problem does not come into one's own.The development applied along with ecommerce etc. in recent years, Face datection becomes
Most potential biometric verification of identity means, this application background requires that automatic face detection system can be to general environment image
Having certain adaptation ability, therefore, Face datection receives the attention of researcher initially as an independent problem.
At present a lot of to Face datection document, under most Main Analysis complex backgrounds multi-angle and also do not reach in real time,
And for the research of driver's Face datection, seldom.The present invention be directed to the Face datection of driver, especially at night
Between or severe driving environment under, by RGB-I photographic head, it is proposed that a kind of driving human pilot people round the clock based on RGB-I
Face real-time detection method, substantially increases accuracy of detection and speed.
Summary of the invention
The problem that the present invention is to solve the detection in real time of driver face round the clock, it is proposed that a kind of daytime based on RGB-I
Vehicle running in the night human pilot face real-time detection method, technical scheme is as follows:
A kind of human pilot face real-time detection method of driving round the clock based on RGB-I, including model training step and
The step of driver's face recognition；
The step of described model training includes:
S1, the pretreatment of driver's face database；
S2, sets up R-Convnet model based on RGB image training and I-based on the training of Infrared image
Convnet model；
The step of described driver's face recognition includes:
S3, uses RGB-Infrared photographic head to obtain driver's image round the clock, by day, obtains driver's in real time
RGB image, obtains the Infrared image of driver in real time at night；
S4, uses the R-Convnet model trained to be identified by day, then uses the I-trained at night
Convnet model is identified；Described R-Convnet model is R40-Convnet model and the R28-Convnet mould of cascade
Type；Described I-Convnet model is I40-Convnet model and the I28-Convnet model of cascade.
Further, in described step S1, pretreatment includes the RGB face database pretreatment of driver and driver
Infrared face database pretreatment；
The RGB face database pretreatment of described driver includes: the RGB face picture of driver's face is divided into two groups: one
The size of picture is adjusted to 40*40 by group, carries out gray processing afterwards；Picture size is adjusted to 28*28 by another group, carries out afterwards
Gray processing；
The Infrared face database pretreatment of described driver includes: the Infrared face picture of driver's face divided
Become two groups: one group the size of picture is adjusted to 40*40, carry out gray processing afterwards and deposit；Picture size is adjusted to 28* by another group
28, carry out gray processing afterwards.
Further, the training method of the R40-Convnet model described in described step S2 is: first to Convnet
Detector carries out parameter setting, and wherein convolution kernel size is respectively 5*5 and 3*3, and convolution kernel quantity is 2 and 5, and the learning efficiency is
0.01, batch size is 200, and frequency of training is 100, the RGB face picture that the pretreatment of step S1 obtains is input to afterwards
In Convnet detector, reach frequency of training, obtain R40-Convnet model；
The training method of described R28-Convnet model is: first Convnet detector is carried out parameter setting, wherein
Convolution kernel size is respectively 5*5 and 3*3, and convolution kernel quantity is 3 and 2, and the learning efficiency is 0.01, and batch size is 100, training
Number of times is 50, the RGB face picture that S1 pretreatment obtains is input in Convnet detector afterwards, reaches frequency of training,
To R28-Convnet model.
Further, the training method of the I40-Convnet model described in described step S2 is: first to Convnet
Detector carries out parameter setting, and wherein convolution kernel size is respectively 5*5 and 3*3, and convolution kernel quantity is 2 and 5, and the learning efficiency is
0.01, batch size is 200, and frequency of training is 100, and the Infrared face picture pretreatment of step S2 obtained afterwards is defeated
Enter in Convnet detector, reach frequency of training, obtain I40-Convnet model.
The training method of described I28-Convnet model is: first Convnet detector is carried out parameter setting, wherein
Convolution kernel size is respectively 5*5 and 3*3, and convolution kernel quantity is 3 and 2, and the learning efficiency is 0.01, and batch size is 100, training
Number of times is 50, the Infrared face picture that the pretreatment of step S2 obtains is input in Convnet detector afterwards, reaches instruction
Practice number of times, obtain I28-Convnet model.
Further, described Convnet detector eliminates full articulamentum on the basis of CNN and obtains.
Further, the realization of described step S4 includes:
By day, to each frame in rgb video, with the window sliding of 40*40, step-length is 2, the figure that sliding window obtains
Carrying out detecting whether face as being input to R40-Convnet model, if face, then the size adjusting this image is 28*28,
It is input to the R28-Convnet model of next cascade, if it is decided that or face, then preserve the position of window this moment, otherwise
Do not preserve, next proceed to sliding window, until after complete test picture of detection, picture being zoomed in and out with the ratio of 0.8,
Repeat above step, until picture ends less than 40*40；Now, overlapping a lot of windows at face location, by non-maximum
Restrainable algorithms is calibrated to a window, i.e. can get driver's face；
At night, to each frame in Infrared video, with the window sliding of 40*40, step-length is 2, and sliding window obtains
To image be input to I40-Convnet model and carry out detecting whether face, if the size that face then adjusts this image is
28*28 continues with next I28-Convnet model, if face, then preserves the position of window this moment, does not protects
Deposit, next proceed to sliding window, until after complete test picture of detection, picture being zoomed in and out with the ratio of 0.8, repeating
Above step, until picture ends less than 40*40；Now, at face location, overlapping a lot of windows, pass through non-maxima suppression
Algorithm is calibrated to a window, i.e. can get driver's face.
Beneficial effects of the present invention:
1, introduce Infrared photographic head, improve only by RGB photographic head at the definition of shooting at night, successfully solve
Driver face real-time test problems round the clock, especially under night or severe driving environment, achieves height by RGB-I
Precision real-time face detects.
2, by the amendment Convnet model that obtains of CNN model, greatly enhance driver's face detection time
Between；
3, the region that sliding window gets, and by the Convnet model of cascade, it is detected whether that face schemes
Picture, improves speed and the accuracy rate of detection.
4, the present invention successfully solves the problem of driver's face detection round the clock not enough (accuracy and real-time), relies on
RGB-I achieves the monitoring of driver's fine definition round the clock, utilizes Convnet detector to detect driver face the most accurately
The detection of expressing one's feelings of the fatigue detecting of driver and driver is extremely important.
Accompanying drawing explanation
Fig. 1 is the flow chart of driver's face detection based on RGB-I image.
Fig. 2 is the schematic diagram of RGB picture training R40-Convnet model based on 40*40.
Fig. 3 is the schematic diagram of RGB picture training R28-Convnet model based on 28*28.
Fig. 4 is the schematic diagram of Infrared picture training I40-Convnet model based on 40*40.
Fig. 5 is the schematic diagram of Infrared picture training I28-Convnet model based on 28*28.
Detailed description of the invention
The invention will be further described with specific embodiment below in conjunction with the accompanying drawings.
Fig. 1 gives the general thought of the present invention, uses RGB-Infrared photographic head to obtain driver's image round the clock,
By day, obtaining the RGB image of driver in real time, for the window sliding of each two field picture 40*40, step-length is 2, sliding window
The R40-Convnet model that the image that mouth obtains is input to train carries out detecting whether face, if face then adjusts this
The size of image is the R28-Convnet model that 28*28 continues with that the next one trains, if it is decided that or face, then
Preserve the position of window this moment, do not preserve, next proceed to sliding window, until after complete test picture of detection, right
Picture zooms in and out with the ratio of 0.8, repeats above step, until picture ends less than 40*40, now, and weight at face location
Having folded a lot of window, non-maxima suppression algorithm to be passed through is calibrated to a window, the driver's face being i.e. identified.
At night, obtaining the Infrared image of driver in real time, that driver's face recognition is used is the I40-Convnet trained
Model and I28-Convnet model, recognition methods is identical with aforementioned method based on RGB image identification driver face.
Fig. 2 to Fig. 5 is R40-Convnet model, R28-Convnet model, I40-Convnet model and I28-respectively
The schematic diagram of Convnet model training.As it can be seen, 4 kinds of Convnet models are similar with CNN model, it is on CNN basis up
Fall full articulamentum.So can improve the speed of detection on the basis of ensureing to carry out accuracy rate, increase the real-time of Face datection
Property.
The present invention is a kind of human pilot face real-time detection method of driving round the clock based on RGB-I, specifically comprises the following steps that
1. driver's face database pretreatment:
The RGB face database pretreatment of 1.1 drivers:
The RGB face picture of driver's face is divided into two groups: one group the size of picture is adjusted to 40*40, laggard
Row gray processing and preserving；Picture size is adjusted to 28*28 by another group, carries out gray processing afterwards and preserves.Purpose is: point
Xun Lian not input R40-Convnet detector and the R28-Convnet detector that input is 28*28 for 40*40 (to be used for detecting
Whether face).
The Infrared face database pretreatment of 1.2 drivers:
The Infrared face picture of driver's face is divided into two groups: one group the size of picture is adjusted to 40*40, it
After carry out gray processing and preserve；Picture size is adjusted to 28*28 by another group, carries out gray processing afterwards and preserves.Purpose
It is: be respectively trained the I40-Convnet detector that input is 40*40 and the I28-Convnet detector that input is 28*28.
2. set up based on RGB image training R-Convnet model and train I-Convnet mould based on Infrared image
Type.
2.1 foundation R40-Convnet model based on RGB image training and R28-Convnet model:
2.1.1 training R40-Convnet model:
Convnet detector eliminates full articulamentum on the basis of CNN and obtains, and model is as in figure 2 it is shown, the most right
Convnet detector carries out parameter setting, and wherein convolution kernel size is respectively 5*5 and 3*3, and convolution kernel quantity is 2 and 5, study
Efficiency is 0.01, and batch size is 200, and frequency of training is 100, and 1.1 pretreatment obtain the RGB face picture of 40*40 afterwards
It is input in model, reaches frequency of training, obtain R40-Convnet model.The numerical value arranged makes the accuracy rate of model the highest.
2.1.2 training R28-Convnet model:
First model as it is shown on figure 3, carry out parameter setting to Convnet detector, and wherein convolution kernel size is respectively 5*5
And 3*3, convolution kernel quantity is 3 and 2, and the learning efficiency is 0.01, and batch size is 100, and frequency of training is 50, pre-by 1.1 afterwards
Process obtains the RGB face picture of 28*28 and is input in model, reaches frequency of training, obtains R28-Convnet model.
2.2 set up based on Infrared image training I40-Convnet model and I28-Convnet model:
2.2.1 training I40-Convnet model:
Model as shown in Figure 4, first carries out parameter setting to Convnet detector, and wherein convolution kernel size is respectively 5*5
And 3*3, convolution kernel quantity is 2 and 5, and the learning efficiency is 0.01, and batch size is 200, and frequency of training is 100, afterwards by 1.2
Pretreatment obtains the Infrared face picture of 40*40 and is input in model, reaches frequency of training, and I40-Convnet model is instructed
Practice.
2.2.2 training I28-Convnet model:
First model as it is shown in figure 5, carry out parameter setting to Convnet detector, and wherein convolution kernel size is respectively 5*5
And 3*3, convolution kernel quantity is 3 and 2, and the learning efficiency is 0.01, and batch size is 100, and frequency of training is 50, pre-by 1.1 afterwards
Process obtains the Infrared face picture of 28*28 and is input in model, reaches frequency of training, I28-Convnet model training
Complete.
3. based on RGB image and the recognition of face of Infrared image:
3.1 R-Convnet Model Identification driver's faces based on RGB image:
To each frame in rgb video, with the window sliding of 40*40, step-length is 2, the image input that sliding window obtains
Carry out detecting whether face to R40-Convnet detector, if it is that 28*28 continuation is defeated that face then adjusts the size of this image
Enter to next R28-Convnet detector, if it is decided that or face, then preserve the position of window this moment, do not preserve,
Next proceed to sliding window, until after complete test picture of detection, picture being zoomed in and out with the ratio of 0.8, repeat with
Upper step, until picture ends less than 40*40, now, overlapping a lot of windows, non-maxima suppression to be passed through at face location
Algorithm is calibrated to a window, i.e. can get driver's face.The method is for the time of daytime driving person's face detection
The biggest lifting is had with accuracy.
3.2 I-Convnet Model Identification driver's faces based on Infrared image:
To each frame in Infrared video, with the window sliding of 40*40, step-length is 2, the image that sliding window obtains
It is input to I40-Convnet detector carry out detecting whether face, if it is that 28*28 continues that face then adjusts the size of this image
Continue and be input to next I28-Convnet detector, if face, then preserve the position of window this moment, do not preserve, connect
Get off to continue sliding window, until after complete test picture of detection, picture being zoomed in and out with the ratio of 0.8, more than repetition
Step, until picture ends less than 40*40, now, overlapping a lot of windows at face location, non-maxima suppression to be passed through is calculated
Method is calibrated to a window, i.e. can get driver's face.The method is for the face detection of nighttime driving person driver
Time and accuracy have the biggest lifting.
The a series of detailed description of those listed above is only for the feasibility embodiment of the present invention specifically
Bright, they also are not used to limit the scope of the invention, all equivalent implementations made without departing from skill of the present invention spirit
Or change should be included within the scope of the present invention.
Claims (6)
1. the human pilot face real-time detection method of driving round the clock based on RGB-I, it is characterised in that include model training
Step and the step of driver's face recognition；
The step of described model training includes:
S1, the pretreatment of driver's face database；
S2, sets up R-Convnet model based on RGB image training and I-Convnet mould based on the training of Infrared image
Type；
The step of described driver's face recognition includes:
S3, uses RGB-Infrared photographic head to obtain driver's image round the clock, by day, obtains the RGB of driver in real time
Image, obtains the Infrared image of driver in real time at night；
S4, uses the R-Convnet model trained to be identified by day, then uses the I-Convnet trained at night
Model is identified；Described R-Convnet model is R40-Convnet model and the R28-Convnet model of cascade；Described I-
Convnet model is I40-Convnet model and the I28-Convnet model of cascade.
A kind of human pilot face real-time detection method of driving round the clock based on RGB-I the most according to claim 1, it is special
Levying and be, in described step S1, pretreatment includes RGB face database pretreatment and the Infrared face database of driver of driver
Pretreatment；
The RGB face database pretreatment of described driver includes: the RGB face picture of driver's face is divided into two groups: one group will
The size of picture is adjusted to 40*40, carries out gray processing afterwards；Picture size is adjusted to 28*28 by another group, carries out gray scale afterwards
Change；
The Infrared face database pretreatment of described driver includes: the Infrared face picture of driver's face is divided into two
Group: the size of picture is adjusted to 40*40 by one group, carries out gray processing afterwards and deposits；Picture size is adjusted to 28*28 by another group,
Carry out gray processing afterwards.
A kind of human pilot face real-time detection method of driving round the clock based on RGB-I the most according to claim 1, it is special
Levying and be, the training method of the R40-Convnet model described in described step S2 is: first carry out Convnet detector
Parameter is arranged, and wherein convolution kernel size is respectively 5*5 and 3*3, and convolution kernel quantity is 2 and 5, and the learning efficiency is 0.01, and batch is big
Little is 200, and frequency of training is 100, afterwards the RGB face picture that the pretreatment of step S1 obtains is input to Convnet detector
In, reach frequency of training, obtain R40-Convnet model；
The training method of described R28-Convnet model is: first Convnet detector is carried out parameter setting, wherein convolution
Core size is respectively 5*5 and 3*3, and convolution kernel quantity is 3 and 2, and the learning efficiency is 0.01, and batch size is 100, frequency of training
It is 50, afterwards the RGB face picture that S1 pretreatment obtains is input in Convnet detector, reaches frequency of training, obtain
R28-Convnet model.
A kind of human pilot face real-time detection method of driving round the clock based on RGB-I the most according to claim 1, it is special
Levying and be, the training method of the I40-Convnet model described in described step S2 is: first carry out Convnet detector
Parameter is arranged, and wherein convolution kernel size is respectively 5*5 and 3*3, and convolution kernel quantity is 2 and 5, and the learning efficiency is 0.01, and batch is big
Little is 200, and frequency of training is 100, and the Infrared face picture that the pretreatment of step S2 obtains is input to Convnet inspection afterwards
Survey in device, reach frequency of training, obtain I40-Convnet model.
The training method of described I28-Convnet model is: first Convnet detector is carried out parameter setting, wherein convolution
Core size is respectively 5*5 and 3*3, and convolution kernel quantity is 3 and 2, and the learning efficiency is 0.01, and batch size is 100, frequency of training
It is 50, afterwards the Infrared face picture that the pretreatment of step S2 obtains is input in Convnet detector, reach training time
Number, obtains I28-Convnet model.
5. according to a kind of based on RGB-I the human pilot face real-time detection method of driving round the clock described in claim 3 or 4,
It is characterized in that, described Convnet detector eliminates full articulamentum on the basis of CNN and obtains.
A kind of human pilot face real-time detection method of driving round the clock based on RGB-I the most according to claim 1, it is special
Levying and be, the realization of described step S4 includes:
By day, to each frame in rgb video, with the window sliding of 40*40, step-length is 2, and the image that sliding window obtains is defeated
Entering and carry out detecting whether face to R40-Convnet model, if face, then the size adjusting this image is 28*28, input
R28-Convnet model to next one cascade, if it is decided that or face, then preserve the position of window this moment, do not protect
Deposit, next proceed to sliding window, until after complete test picture of detection, picture being zoomed in and out with the ratio of 0.8, repeating
Above step, until picture ends less than 40*40；Now, at face location, overlapping a lot of windows, pass through non-maxima suppression
Algorithm is calibrated to a window, i.e. can get driver's face；
At night, to each frame in Infrared video, with the window sliding of 40*40, step-length is 2, and sliding window obtains
Image is input to I40-Convnet model to carry out detecting whether face, if it is 28*28 that face then adjusts the size of this image
Continue with next I28-Convnet model, if face, then preserve the position of window this moment, do not preserve, connect
Get off to continue sliding window, until after complete test picture of detection, picture being zoomed in and out with the ratio of 0.8, more than repetition
Step, until picture ends less than 40*40；Now, overlapping a lot of windows at face location, by non-maxima suppression algorithm
It is calibrated to a window, i.e. be can get driver's face.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610436625.1A CN106127123B (en) | 2016-06-16 | 2016-06-16 | Method for detecting face of driver in day and night driving in real time based on RGB-I |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610436625.1A CN106127123B (en) | 2016-06-16 | 2016-06-16 | Method for detecting face of driver in day and night driving in real time based on RGB-I |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106127123A true CN106127123A (en) | 2016-11-16 |
| CN106127123B CN106127123B (en) | 2019-12-31 |
Family
ID=57471063
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610436625.1A Active CN106127123B (en) | 2016-06-16 | 2016-06-16 | Method for detecting face of driver in day and night driving in real time based on RGB-I |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106127123B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107729986A (en) * | 2017-09-19 | 2018-02-23 | 平安科技（深圳）有限公司 | Driving model training method, driver's recognition methods, device, equipment and medium |
| CN109271947A (en) * | 2018-09-28 | 2019-01-25 | 合肥工业大学 | A kind of night real-time hand language identifying system based on thermal imaging |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060102843A1 (en) * | 2004-11-12 | 2006-05-18 | Bazakos Michael E | Infrared and visible fusion face recognition system |
| US20060104488A1 (en) * | 2004-11-12 | 2006-05-18 | Bazakos Michael E | Infrared face detection and recognition system |
| JP2008305203A (en) * | 2007-06-07 | 2008-12-18 | Denso Corp | Face imaging device |
| CN203193791U (en) * | 2013-03-26 | 2013-09-11 | 深圳市中控生物识别技术有限公司 | Network camera with face identification function |
| CN103496341A (en) * | 2013-10-10 | 2014-01-08 | 扬州瑞控汽车电子有限公司 | Vehicle-mounted infrared night-vision imaging system based on vehicle active safety |
| WO2014176485A1 (en) * | 2013-04-26 | 2014-10-30 | West Virginia High Technology Consortium Foundation, Inc. | Facial recognition method and apparatus |
| CN104318237A (en) * | 2014-10-28 | 2015-01-28 | 厦门大学 | Fatigue driving warning method based on face identification |
| CN104580896A (en) * | 2014-12-25 | 2015-04-29 | 深圳市锐明视讯技术有限公司 | Method and device for switching day and night modes of camera |
| CN105389546A (en) * | 2015-10-22 | 2016-03-09 | 四川膨旭科技有限公司 | System for identifying person at night during vehicle driving process |
-
2016
- 2016-06-16 CN CN201610436625.1A patent/CN106127123B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060102843A1 (en) * | 2004-11-12 | 2006-05-18 | Bazakos Michael E | Infrared and visible fusion face recognition system |
| US20060104488A1 (en) * | 2004-11-12 | 2006-05-18 | Bazakos Michael E | Infrared face detection and recognition system |
| JP2008305203A (en) * | 2007-06-07 | 2008-12-18 | Denso Corp | Face imaging device |
| CN203193791U (en) * | 2013-03-26 | 2013-09-11 | 深圳市中控生物识别技术有限公司 | Network camera with face identification function |
| WO2014176485A1 (en) * | 2013-04-26 | 2014-10-30 | West Virginia High Technology Consortium Foundation, Inc. | Facial recognition method and apparatus |
| CN103496341A (en) * | 2013-10-10 | 2014-01-08 | 扬州瑞控汽车电子有限公司 | Vehicle-mounted infrared night-vision imaging system based on vehicle active safety |
| CN104318237A (en) * | 2014-10-28 | 2015-01-28 | 厦门大学 | Fatigue driving warning method based on face identification |
| CN104580896A (en) * | 2014-12-25 | 2015-04-29 | 深圳市锐明视讯技术有限公司 | Method and device for switching day and night modes of camera |
| CN105389546A (en) * | 2015-10-22 | 2016-03-09 | 四川膨旭科技有限公司 | System for identifying person at night during vehicle driving process |
Non-Patent Citations (3)
| Title |
|---|
| Y. YOSHITOMI等: "Facial expression recognition using thermal image processing and neural network", 《PROCEEDINGS 6TH IEEE INTERNATIONAL WORKSHOP ON ROBOT AND HUMAN COMMUNICATION》 * |
| 汪济民: "基于卷积神经网络的人脸检测和性别识别研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 蔡菁 等: "基于Kinect的疲劳驾驶综合监测***研究", 《计算机工程与科学》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107729986A (en) * | 2017-09-19 | 2018-02-23 | 平安科技（深圳）有限公司 | Driving model training method, driver's recognition methods, device, equipment and medium |
| CN109271947A (en) * | 2018-09-28 | 2019-01-25 | 合肥工业大学 | A kind of night real-time hand language identifying system based on thermal imaging |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106127123B (en) | 2019-12-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106407931B (en) | A kind of depth convolutional neural networks moving vehicle detection method | |
| CN106611169B (en) | A kind of dangerous driving behavior real-time detection method based on deep learning | |
| CN108446678B (en) | Dangerous driving behavior identification method based on skeletal features | |
| CN106803069A (en) | Crowd's level of happiness recognition methods based on deep learning | |
| CN108609019A (en) | A kind of electric vehicle automatic Pilot method based on artificial intelligence platform | |
| CN108830252A (en) | A kind of convolutional neural networks human motion recognition method of amalgamation of global space-time characteristic | |
| CN107491769A (en) | Method for detecting fatigue driving and system based on AdaBoost algorithms | |
| CN105608446A (en) | Video stream abnormal event detection method and apparatus | |
| CN106845327B (en) | Training method, face alignment method and the device of face alignment model | |
| CN110147738B (en) | Driver fatigue monitoring and early warning method and system | |
| CN105892471A (en) | Automatic automobile driving method and device | |
| CN109919938B (en) | Method for obtaining optic disc segmentation atlas of glaucoma | |
| CN110728241A (en) | Driver fatigue detection method based on deep learning multi-feature fusion | |
| Yu et al. | Railway obstacle detection algorithm using neural network | |
| CN110956082B (en) | Face key point detection method and detection system based on deep learning | |
| CN110532925B (en) | Driver fatigue detection method based on space-time graph convolutional network | |
| CN107203752A (en) | A kind of combined depth study and the face identification method of the norm constraint of feature two | |
| CN109034134A (en) | Abnormal driving behavioral value method based on multitask depth convolutional neural networks | |
| CN110569843A (en) | Intelligent detection and identification method for mine target | |
| CN112749675A (en) | Potato disease identification method based on convolutional neural network | |
| CN104077568A (en) | High-accuracy driver behavior recognition and monitoring method and system | |
| CN107633196A (en) | A kind of eyeball moving projection scheme based on convolutional neural networks | |
| Nose et al. | A study on a lane keeping system using CNN for online learning of steering control from real time images | |
| CN110276398A (en) | A kind of video abnormal behaviour automatic judging method | |
| CN106127123A (en) | A kind of human pilot face real-time detection method of driving round the clock based on RGB I |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |