CN109389092A - A kind of local enhancement multitask depth migration transfinites the facial video fatigue detection method of learning machine and individual robust - Google Patents

A kind of local enhancement multitask depth migration transfinites the facial video fatigue detection method of learning machine and individual robust Download PDF

Info

Publication number
CN109389092A
CN109389092A CN201811232518.2A CN201811232518A CN109389092A CN 109389092 A CN109389092 A CN 109389092A CN 201811232518 A CN201811232518 A CN 201811232518A CN 109389092 A CN109389092 A CN 109389092A
Authority
CN
China
Prior art keywords
learning machine
task
subtask
transfinites
data
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
Application number
CN201811232518.2A
Other languages
Chinese (zh)
Other versions
CN109389092B (en
Inventor
贾熹滨
李威庭
王悦宸
苏醒
郭黎敏
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing University of Technology
Original Assignee
Beijing University of Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Beijing University of Technology filed Critical Beijing University of Technology
Priority to CN201811232518.2A priority Critical patent/CN109389092B/en
Publication of CN109389092A publication Critical patent/CN109389092A/en
Application granted granted Critical
Publication of CN109389092B publication Critical patent/CN109389092B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • 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
    • G06V40/174—Facial expression recognition
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Multimedia (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Human Computer Interaction (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

It transfinites the facial video fatigue detection method of learning machine and individual robust the invention discloses a kind of local enhancement multitask depth migration, this method includes the acquisition and pretreatment of data, acquires initial data with common camera；The sound state binding characteristic of data indicates；Sound state Fusion Features are carried out to data, extract image sequence static nature and behavioral characteristics respectively, then behavioral characteristics and static nature are carried out with the fusion in geometry level.It transfinites in learning machine in depth with multi-task mechanism, main task will be set as about fatigue detecting, other face recognition types are set as subtask, pass through main task, subtask is separately trained, and identification is merged, and removes influence of the subtask to main task, network is improved to the robustness of different people fatigue detecting, the demand to different personnel's robusts has been reached in the case where guaranteeing discrimination.

Description

A kind of local enhancement multitask depth migration transfinites the face of learning machine and individual robust Video fatigue detection method
Technical field
The present invention relates to computer vision, pattern-recognition, the computer application fields such as image procossing, and in particular to a benefit The method that the public effective coverage of facial video data is extracted with the learning machine that transfinites-autocoder reconstruction result, a utilization The depth that the public effective coverage carries out regional area enhancing transfinites learning machine method；One is added multi-task mechanism reality wherein Now to the method for non-trainer's fatigue detecting；And it is super using the above-mentioned local enhancement multitask depth of transfer learning thought training The method for limiting learning machine.
Background technique
Fatigue is a kind of to appear in psychology and state physiologically.Mental fatigue refer to continuously be engaged in for a long time it is bored, The weary state of spirit caused by dull or prolonged mental labour and maintenance high-pressure state.Mental fatigue usually shows To be in a disturbed state of mind, reaction speed slows down, and motivation is lost.Physiological fatigue is then physiology caused by being worked due to prolonged Non-intermittent Variation, is usually expressed as uncomfortable impression caused by the accumulation of metabolic waste in the body, including but not limited to lactic acid The accumulation of the decline of muscle tone caused by being accumulated in muscle and the reduction of movement durability degree and carbon dioxide in respiratory system It is caused to yawn.
The local enhancement learning machine that transfinites improves contribution degree of the important regional area in identification, to improve identification essence Degree.The local strengthening depth learning machine that transfinites is broadly divided into two steps, detects important regional area first, then passes through training Local strengthening transfinites learning machine self-encoding encoder, successively stacks self-encoding encoder and forms depth network structure.The depth of local strengthening is super Limiting learning machine has more complicated structure and richer information.
Hiding node layer is divided into main task, two, subtask part by multitask extreme learning machine.Main task part it is hidden It is identical as ELM to hide node layer calculation method.Subtask number herein is 8 unsigned numbers, range 1-255, maximum support 255 subtasks.Main task output par, c is that acceptor task number does not influence, and subtask offset portion is in the training process The part that acceptor task number influences.In the training stage, with the mark of the sum of main task output and subtask biasing fitting training data Label, training β matrix.In cognitive phase, remove subtask offset portion, only calculates estimating for output vector with main task output par, c Evaluation improves main task recognition effect to realize the influence of removal subtask.It is usually relatively small, because in training In stage output vector estimated value, main task output par, c plays a major role, and subtask offset portion plays a secondary role.In we In method, multitask extreme learning machine can remove the influence of different experiments subjects face movement otherness, improve to fatigue state Detection effect.
The transfer of learning refers to a kind of experience of influence or acquistion for learning to learn another kind to other movable shadows of completion It rings.Migration is widely present in the study of various knowledge, technical ability and social regulation.Due to learning activities be always built upon it is existing On knowledge experience, this process that new knowledge and technical ability are constantly obtained using existing knowledge experience, it is believed that be The transfer of learning of broad sense；And it is rare due to fatigue detecting sets of video data herein, using public facial expression data collection CAS (ME), the data sets such as JAFEE, CK training it is public face enhancing region extractor (learning machine-autocoder that transfinites) with And face feature extractor (stacking transfinite learning machine-autocoder).The parameter trained, which is fixed, directly to be applied to Among fatigue detecting task, and fatigue state point is only really eventually used for using adopt that fatigue detecting data set is trained certainly The multitask of class is transfinited learning machine.
Summary of the invention
The purpose of the present invention is to provide a kind of, and the driving fatigue based on facial video individual robust detects method for early warning.It should Method should have data easily to acquire, and use cost is low, real-time detection early warning, and have certain identification to the personnel not trained The features such as rate.
To achieve the goals above, the invention adopts the following technical scheme: acquiring original video from common camera.To original Beginning video by KLT face tracking algorithm carry out facial area tracking, and to the face-image sequence tracked carry out cut and The operation such as size normalization and gray processing.Obtained facial area image sequence, which is carried out sound state binding characteristic, to be indicated, wherein Static nature is to choose the frame image that expression is most plentiful in facial area sequence, and behavioral characteristics use MHI (motion history Figure) algorithm.It the feature of fusion is finally input to trained region enhancing multitask depth transfinites and carry out fatigue in learning machine The detection of state, and will test result and be output to interactive interface to user's progress early warning.Region enhancing multitask depth transfinites Habit machine is that transfinite to the depth input of learning machine carries out region enhancing, and carries out multitask extension, same in order to improve training speed Shi Tigao personnel's universality transfinites learning machine with the thought training region enhancing multitask depth of transfer learning.
The fatigue detection method of personal robust based on facial video, it is characterised in that: realize from common camera to Fatigue state real-time end to end monitoring to the intuitive visible fatigue state recognition result of user, and not to test object And have stronger robustness method includes the following steps:
Step 1, facial acquisition and pretreatment.
Step 1.1, one second 25 frame image is continuously acquired from camera.
Step 1.2, the facial area position of every frame image is determined according to face tracking algorithm.
Step 1.3, the cutting that facial area is carried out to the original image of acquisition, by facial area image scaling to 232* 191 realize the normalization of face-image.
Step 1.4, gray processing is carried out to the facial area image sequence after normalization, reduces data dimension.
Step 2, the sound state combined data of face-image sequence indicates.
Step 2.1, pre-identification is carried out to facial image sequence, chooses its expression and shows the most abundant frame image.
Step 2.2, the behavioral characteristics of face-image sequence are extracted using MHI (motion history figure).
Step 2.3, the static frames of step 2.1 are subjected to merging for geometrical layers with the MHI behavioral characteristics of step 2.2.
Step 3, the inspection of fatigue state is carried out using the trained depth feature vector that learning machine obtains step 2 that transfinites It surveys.
Step 3.1, feature vector step 2 obtained, which is input to trained local enhancement multitask Transfer Depth, to transfinite In learning machine, output vector is calculated.
Step 3.2, to output vector carry out fatigue state judgement, by judging result be output to system interface to user into Row prompt.
Compared with prior art, the present invention has following clear superiority:
The present invention, which is combined, acquires initial data with common camera, and data acquisition cost is low.Sound state knot is carried out to data The character representation of conjunction, it is relatively good to the expressive force of data, and the time of feature extraction is not grown.It is adopted in the context of detection of fatigue state It is detected with the depth of the fast speed learning machine that transfinites.And it transfinites in this depth and facial area is carried out to input data on learning machine Domain enhancing improves the expressive force of facial video.It transfinites simultaneously in depth and multi-task mechanism is added in learning machine, improve to not training Cross the fatigue state discrimination of personnel.The rare of facial video fatigue detecting data set is considered simultaneously, in the training to network The middle thought with migration improves network to the universality of face.This method substantially conforms to market and detects early warning to driving fatigue The use demand of system.
Detailed description of the invention
Fig. 1 is general module design drawing of the invention；
Fig. 2 is facial video acquisition and pretreatment process figure；
The region Fig. 3 enhancing multitask depth transfinites learning machine network；
Specific embodiment
Yi Xiajiehejutishishili,Bing Canzhaofutu,Dui Benfamingjinyibuxiangxishuoming.
The general module of the method for the invention as shown in Figure 1, specifically includes the following steps:
Facial video sequence is obtained first from common camera.Then the cutting of facial area is carried out to it, size is returned One changes, color gray processing.It extracts the dynamic static nature of facial area image sequence respectively again, then feature is carried out in geometry level Fusion.It fusion feature is finally input to trained local enhancement multitask depth transfinites and carry out fatigue state in learning machine Detection.
And the acquisition of facial video therein and its pretreatment process are as shown in Figure 2.
Local enhancement multitask depth proposed by the present invention transfinite learning machine network structure as shown in figure 3, its have with Lower feature:
It transfinites in depth and the regional area based on adaptation mechanism is added in learning machine enhances to improve Network Recognition rate.Add Enter multi-task mechanism, enhancing network detects discrimination to the fatigue state for the people not trained.It applies and moves in training simultaneously The thought of shifting reduces the training time.
Local enhancement methods, are originally inputted using autocoder simulation, and obtained reconstruction result is done with being originally inputted It is done again after XOR operation and operation, obtains the public area for being all able to retain in reconstruct jointly for pictures all in data set Domain.By the pixel value of public domain be originally inputted input more for merge to obtain new region and enhancing carried out in geometry level Hiding node layer is divided into main task, two, subtask part by business mechanism.The hiding node layer calculation method of main task part with ELM is identical.This method neutron task number is 8 unsigned numbers, range 1-255,255 subtasks of maximum support.It is (necessary When can also use 16 unsigned numbers instead, maximum supports 65535 subtasks.) node each node in subtask has 8 groups to join at random Number, every group of parameter include the vector a that length is input vector length n1j...a8jWith real number b1j...b8j, wherein j is subtask section The number of point.If certain training data subtask number is task, the calculating of the weight vectors a and biasing b of subtask node are such as Under:
Wherein (task& (1<<(k-1)))>>(k-1) part indicates the kth position of subtask number (from low level to a high position) Value.If kth position is 1, which is 1；If kth position is 0, which is 0.All random numbers of this method all (0, 1) section, as long as therefore subtask task non-zero, according to this formula calculate weight vectors aj(task) all dimensions and Bias bjIt (task) is all (0,1) section.
Subtask hides the calculation method of node layer are as follows:
hij=g (aj(taski)xi+bj(taski)) (3)
Identification process neutron task number task=0.Then weight vectors a is null vectorBias b=0.If activation primitive is Sigmoid function, then hidden layer nodal value in subtask is 0.5.
Using formula (1), the calculating weight vectors a of (2) and biasing b, the purpose is to the training datas to different subtasks Using different weight vectors and biasing, to achieve the purpose that influence subtask concealed nodes acceptor task number.Weight to The dimension for measuring a is very high, therefore in order to support more subtask, cannot for each subtask concealed nodes every kind of subtask with Machine generates one group of a, b parameter.255 can be supported using 8 groups of random parameters a, b using calculation method shown in formula 1,2 More subtasks to save EMS memory occupation, or are supported in the case where same memory occupies in subtask.Main task it is hidden It is identical as the hiding node layer calculation method of ELM to hide node layer.
The local enhancement multitask learning machine that transfinites is divided into three parts, be first extract enhancing region transfinite learning machine-from Dynamic encoder and the trained stacking of learning machine efficient training method that transfinites utilize to transfinite learning machine-autocoder and for classifying Multitask extension transfinite learning machine.Preceding part transfinite learning machine-autocoder and stacking transfinites learning machine-autocoding The purpose of device is to extract to the public domain for increasing discrimination and automatically extract the feature being originally inputted respectively.Its needs is huge Data volume, and with specific classification task without very big decisive role.And multitask of the latter half for classification is super It is then closely bound up with fatigue detecting task to limit learning machine, but due to currently without the facial sets of video data for fatigue detecting, And the acquisition of data set is adopted certainly and is manually marked and needs many manpower and material resources.
In this context, this method is transfinited learning machine using the thought training local enhancement multitask depth of migration.It is walked It is rapid as follows:
Step 1, at CAS (ME), the face video common data sets such as JAFEE, JK, a large amount of facial video data is utilized Learning machine-the autocoder that transfinites is trained, is extracted using the distribution of reconfigured geometry and all data is all protected in restructuring procedure The public domain stayed.
Step 2, using above-mentioned public domain, extract CAS (ME), the pixel value of the corresponding region of JAFEE, JK, by its with Initial data carries out the fusion in geometry level, obtains new input data.
Step 3, the learning machine-autocoder that transfinites is stacked using new input data training, makes it have and extracts newly Input number
According to the ability of feature.
Transfinite learning machine-autocoder parameter for the public domain and stacking that reservation above-mentioned steps 2 and step 3 obtain. From the data set for fatigue detecting adopted, the pixel value for extracting public enhancing region is several with raw data set progress by it for input The fusion of what layer, then be entered into stacking and transfinite and extract feature in learning machine-autocoder.Utilize obtained feature training Multitask for fatigue detecting task is transfinited learning machine.Fixed obtained parameter is for last fatigue detecting early warning task
So far, specific implementation process of the invention is just described.

Claims (5)

  1. The facial video fatigue detection method of learning machine and individual robust 1. a kind of local enhancement multitask depth migration transfinites, It is characterized in that: being input with the user's face video that camera acquires, realize real-time fatigue state monitoring end to end, promotion pair The adaptability of user object individual difference, method includes the following steps:
    Step 1, facial acquisition and pretreatment；
    Step 1.1, one second 25 frame image is continuously acquired from camera；
    Step 1.2, the facial area position of every frame image is determined according to face tracking algorithm；
    Step 1.3, its area image is zoomed to 232*191 and realized by the cutting that facial area is carried out to the original image of acquisition The normalization of face-image；
    Step 1.4, gray processing is carried out to the facial area image sequence after normalization, reduces data dimension；
    Step 2, the sound state combined data of face-image sequence indicates；
    Step 2.1, pre-identification is carried out to facial image sequence, chooses its expression and shows the most abundant frame image；
    Step 2.2, the behavioral characteristics of face-image sequence are extracted using MHI, that is, motion history figure；
    Step 2.3, the static frames of step 2.1 are subjected to merging for geometrical layers with the MHI behavioral characteristics of step 2.2；
    Step 3, the detection of fatigue state is carried out using the trained depth feature vector that learning machine obtains step 2 that transfinites；
    Step 3.1, feature vector step 2 obtained, which is input to trained local enhancement multitask Transfer Depth, to transfinite study In machine, output vector is calculated；
    Step 3.2, the judgement that fatigue state is carried out to output vector, is output to system interface for judging result and mentions to user Show.
  2. The face of learning machine and individual robust 2. a kind of local enhancement multitask depth migration according to claim 1 transfinites Video fatigue detection method, it is characterised in that: transfinite in depth and the regional area based on adaptation mechanism is added in learning machine increases By force to improve Network Recognition rate；Multi-task mechanism is added, enhancing network detects discrimination to the fatigue state for the people not trained； The thought for applying migration in training simultaneously, reduces the training time.
  3. The face of learning machine and individual robust 3. a kind of local enhancement multitask depth migration according to claim 2 transfinites Video fatigue detection method, it is characterised in that: using autocoder simulation be originally inputted, by obtained reconstruction result with it is original Input is done again after doing XOR operation and operation, obtains the public affairs for being all able to retain in reconstruct jointly for pictures all in data set Region altogether；By the pixel value of public domain be originally inputted carry out in geometry level merge to obtain new region enhance it is defeated Enter.
  4. The face of learning machine and individual robust 4. a kind of local enhancement multitask depth migration according to claim 2 transfinites Video fatigue detection method, it is characterised in that: hiding node layer is divided into main task, subtask two by the multi-task mechanism A part；The hiding node layer calculation method of main task part is identical as ELM；This method neutron task number is one 8 without symbol Number, range 1-255, maximum support 255 subtasks；Node each node in subtask has 8 groups of random parameters, every group of parameter packet Include the vector a that length is input vector length n1j...a8jWith real number b1j…b8j, wherein j is the number of subtask node；If Certain training data subtask number is task, then the calculation method of the weight vectors a of subtask node and biasing b are as follows:
    Wherein (task& (1<<(k-1)))>>part (k-1) indicate subtask number kth position value, k is from low level to a high position；Such as Fruit
    Kth position is 1, then the part is 1；If kth position is 0, which is 0；All random numbers of this method are all in (0,1)
    Section, as long as therefore subtask task non-zero, according to this formula calculate weight vectors aj(task) all dimensions and
    Bias bjIt (task) is all (0,1) section；
    Subtask hides the calculation method of node layer are as follows:
    hij=g (aj(taski)xi+bj(taski)) (3)
    Identification process neutron task number task=0；Then weight vectors a is null vectorBias b=0；If activation primitive is Sigmoid function, then hidden layer nodal value in subtask is 0.5；
    Formula (1), (2) has been used to calculate weight vectors a and biasing b, the purpose is to the training data uses to different subtasks Different weight vectors and biasing, to achieve the purpose that influence subtask concealed nodes acceptor task number；Weight vectors a's Dimension is very high, therefore in order to support more subtask, cannot give birth at random for every kind of subtask of each subtask concealed nodes At one group of a, b parameter；255 subtasks can be supported, to save using 8 groups of random parameters a, b using the calculating of formula 1,2 About EMS memory occupation, or more subtasks are supported in the case where same memory occupies；The hiding node layer and ELM of main task Hiding node layer calculation method it is identical.
  5. The face of learning machine and individual robust 5. a kind of local enhancement multitask depth migration according to claim 2 transfinites Video fatigue detection method, it is characterised in that: it is transfinited learning machine using the thought training local enhancement multitask depth of migration, office Portion's enhancing multitask learning machine that transfinites is divided into three parts, and its step are as follows:
    Step 1, at CAS (ME), JAFEE is super using a large amount of facial video data training under JK face video common data sets Learning machine-autocoder is limited, extracts the public affairs all retained in restructuring procedure all data using the distribution of reconfigured geometry Region altogether；
    Step 2, using above-mentioned public domain, extract CAS (ME), the pixel value of the corresponding region of JAFEE, JK, by its with it is original Data carry out the fusion in geometry level, obtain new input data；The learning machine-that transfinites is stacked using new input data training Autocoder makes it have the ability for extracting new input data feature；
    Step 3, transfinite learning machine-autocoder ginseng for the public domain and stacking that reservation above-mentioned steps 2 and step 3 obtain Number；Input from the data set for fatigue detecting adopted, extract the pixel value in public enhancing region by itself and raw data set into The fusion of row geometrical layers, then be entered into stacking and transfinite and extract feature in learning machine-autocoder；Utilize obtained feature Multitask of the training for fatigue detecting task is transfinited learning machine；Fixed obtained parameter is appointed for last fatigue detecting early warning Business.
CN201811232518.2A 2018-10-22 2018-10-22 Locally-enhanced multitasking depth migration overrun learning machine and individual robust facial video fatigue detection method Active CN109389092B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811232518.2A CN109389092B (en) 2018-10-22 2018-10-22 Locally-enhanced multitasking depth migration overrun learning machine and individual robust facial video fatigue detection method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811232518.2A CN109389092B (en) 2018-10-22 2018-10-22 Locally-enhanced multitasking depth migration overrun learning machine and individual robust facial video fatigue detection method

Publications (2)

Publication Number Publication Date
CN109389092A true CN109389092A (en) 2019-02-26
CN109389092B CN109389092B (en) 2023-05-02

Family

ID=65427841

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811232518.2A Active CN109389092B (en) 2018-10-22 2018-10-22 Locally-enhanced multitasking depth migration overrun learning machine and individual robust facial video fatigue detection method

Country Status (1)

Country Link
CN (1) CN109389092B (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100329517A1 (en) * 2009-06-26 2010-12-30 Microsoft Corporation Boosted face verification
CN105354988A (en) * 2015-12-11 2016-02-24 东北大学 Driver fatigue driving detection system based on machine vision and detection method
CN106446849A (en) * 2016-09-30 2017-02-22 防城港市港口区高创信息技术有限公司 Fatigue driving detection method
CN106960176A (en) * 2017-02-22 2017-07-18 华侨大学 A kind of pedestrian's gender identification method based on transfinite learning machine and color characteristic fusion
CN107451651A (en) * 2017-07-28 2017-12-08 杭州电子科技大学 A kind of driving fatigue detection method of the H ELM based on particle group optimizing

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100329517A1 (en) * 2009-06-26 2010-12-30 Microsoft Corporation Boosted face verification
CN105354988A (en) * 2015-12-11 2016-02-24 东北大学 Driver fatigue driving detection system based on machine vision and detection method
CN106446849A (en) * 2016-09-30 2017-02-22 防城港市港口区高创信息技术有限公司 Fatigue driving detection method
CN106960176A (en) * 2017-02-22 2017-07-18 华侨大学 A kind of pedestrian's gender identification method based on transfinite learning machine and color characteristic fusion
CN107451651A (en) * 2017-07-28 2017-12-08 杭州电子科技大学 A kind of driving fatigue detection method of the H ELM based on particle group optimizing

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
赵旭阳 等: "基于极限学习机面部识别的疲劳驾驶监测方法的研究", 《中国设备工程》 *

Also Published As

Publication number Publication date
CN109389092B (en) 2023-05-02

Similar Documents

Publication Publication Date Title
CN109583342B (en) Human face living body detection method based on transfer learning
US11908240B2 (en) Micro-expression recognition method based on multi-scale spatiotemporal feature neural network
CN105718873B (en) Stream of people&#39;s analysis method based on binocular vision
CN110464366A (en) A kind of Emotion identification method, system and storage medium
Tang et al. View-independent facial action unit detection
CN104143079A (en) Method and system for face attribute recognition
CN111027377B (en) Double-flow neural network time sequence action positioning method
CN110765873A (en) Facial expression recognition method and device based on expression intensity label distribution
CN110163567A (en) Classroom roll calling system based on multitask concatenated convolutional neural network
CN110929687A (en) Multi-user behavior recognition system based on key point detection and working method
CN113869276B (en) Lie recognition method and system based on micro-expression
CN110414367A (en) A kind of timing behavioral value method based on GAN and SSN
CN108985223A (en) A kind of human motion recognition method
CN111814713A (en) Expression recognition method based on BN parameter transfer learning
Rahul et al. A new hybrid approach for efficient emotion recognition using deep learning
Narayanan et al. Yoga pose detection using deep learning techniques
Sulong et al. HUMAN ACTIVITIES RECOGNITION VIA FEATURES EXTRACTION FROM SKELETON.
Sakthimohan et al. Detection and Recognition of Face Using Deep Learning
CN114937298A (en) Micro-expression recognition method based on feature decoupling
Pang et al. Dance video motion recognition based on computer vision and image processing
Sujatha et al. Enhancing Object Detection with Mask R-CNN: A Deep Learning Perspective
CN103310454B (en) Stationary object type in retentate detection judges to analyze method and system with owner
CN110826459B (en) Migratable campus violent behavior video identification method based on attitude estimation
CN113705301A (en) Image processing method and device
CN109389092A (en) A kind of local enhancement multitask depth migration transfinites the facial video fatigue detection method of learning machine and individual robust

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
GR01 Patent grant
GR01 Patent grant