CN106778528A - A kind of method for detecting fatigue driving based on gaussian pyramid feature - Google Patents

A kind of method for detecting fatigue driving based on gaussian pyramid feature Download PDF

Info

Publication number
CN106778528A
CN106778528A CN201611062106.XA CN201611062106A CN106778528A CN 106778528 A CN106778528 A CN 106778528A CN 201611062106 A CN201611062106 A CN 201611062106A CN 106778528 A CN106778528 A CN 106778528A
Authority
CN
China
Prior art keywords
carried out
feature
fatigue driving
gaussian pyramid
gaussian
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201611062106.XA
Other languages
Chinese (zh)
Inventor
张卫华
周激流
周琳琳
张意
林峰
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.)
Sichuan University
Original Assignee
Sichuan University
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 Sichuan University filed Critical Sichuan University
Priority to CN201611062106.XA priority Critical patent/CN106778528A/en
Publication of CN106778528A publication Critical patent/CN106778528A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/59—Context or environment of the image inside of a vehicle, e.g. relating to seat occupancy, driver state or inner lighting conditions
    • G06V20/597—Recognising the driver's state or behaviour, e.g. attention or drowsiness
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
    • G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
    • G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]

Abstract

The invention discloses a kind of method for detecting fatigue driving based on gaussian pyramid feature, comprise the following steps：S1, the driver's driving image to shooting are pre-processed；S2, the pretreated view data is carried out down-sampled, obtain multi-resolution Gaussian pyramid feature；S3, the feature is carried out to be matched with the property data base for prestoring, judge whether driver is fatigue driving.Method for detecting fatigue driving based on gaussian pyramid feature of the invention uses gaussian pyramid characteristic analysis method, and algorithm complex is low compared to existing technology, computational efficiency is high, recognition accuracy is high.

Description

A kind of method for detecting fatigue driving based on gaussian pyramid feature
Technical field
The present invention relates to field of image recognition, more particularly to a kind of fatigue driving detection side based on gaussian pyramid feature Method.
Background technology
With the development of social economy, automobile turns into the core power of transport article and traffic in daily life Amount, unfortunately with the growth of automobile usage amount, causes to cause the number of dead and grievous injury to increase year by year because of traffic accident Plus.Therefore, traffic safety problem has attracted increasing attention, and improves vehicle safety performance, reduces road traffic accident It is the social concern that merits attention and in the urgent need to the important topic solved by technological meanses.Row of the active safe driving with people It is and custom is closely related.Driver behavior is monitored and is recognized, can in advance be analyzed the security of driving behavior, reminded not Safety behavior is avoiding the generation of accident.In driving procedure, operate the hand foot action and human pilot head of automobile dynamic Work is the direct expression of driving behavior, such as checks instrument board, observation rearview mirror etc.；But there is also some actions unrelated with driving TR is such as seen, watched attentively outside side window for a long time, phoned with mobile telephone in long-time operation vehicle electronic device, startup procedure and bury Head lost article found etc., this class behavior generates great hidden danger to driving safety.Therefore, develop one kind and ensure algorithm real-time as far as possible Under the conditions of, a small amount of quick acting recognizer for sacrificing discrimination, another thinking as action recognition research.Motion is detection The important information of target and background segment in scene, such method assumes initially that the periodicity of pedestrian's action, according to this hypothesis Pedestrian is come with background separation.Human action identification is challenging problem in computer vision field, outside different people The change of the difference of sight, unstable background, the video camera of movement and scene illumination all adds difficulty to action recognition.Tradition Vision sensor under the conditions of, human action can be showed with the consecutive variations between consecutive frame in video sequence.Analysis method It is broadly divided into three classes：The first kind is that (Motion Histogram Image move Nogata to the signature analysis such as MHI based on space-time Figure), SI (Silhouette Histogram profiles histogram) and optical flow analysis and space-time characteristic body etc.；Equations of The Second Kind is to human body Trunk is modeled with four limbs, and then human body in each two field picture in video sequence is detected and model parameter is obtained to retouch State action change；3rd class carries out statistics identification, such as research of Ke Yan to frame of video bottom-up information with the method for image statisticses In, image in video sequence is carried out into color segmentation, then enter action using the statistical information of same color block in consecutive image Recognize.
However, this kind of method computational complexity of signature analysis that the first kind is based on space-time is low, be easily achieved, but image is made an uproar Sound is more sensitive；The action description accuracy that Equations of The Second Kind is modeled the method realization to trunk and four limbs is good, but amount of calculation Greatly, real-time is relatively low；3rd class carries out statistics identification with the method for image statisticses to frame of video bottom-up information, can not meet in real time Property and the requirement being easily achieved.In sum, traditional recognizer computational complexity is high, efficiency is low；Recognition accuracy is low.
The content of the invention
The invention reside in the above-mentioned deficiency for overcoming prior art, there is provided one kind can reduce algorithm complex, computational efficiency High, the recognition accuracy method for detecting fatigue driving based on gaussian pyramid feature high.
In order to realize foregoing invention purpose, the technical solution adopted by the present invention is：
A kind of method for detecting fatigue driving based on gaussian pyramid feature, comprises the following steps：
S1, the driver's driving image to shooting are pre-processed；
S2, the pretreated view data is carried out down-sampled, obtain multi-resolution Gaussian pyramid feature；
S3, the feature is carried out to be matched with the property data base for prestoring, judge whether driver is fatigue Drive.
Further, the step S1 includes, light stream, Gaussian smoothing, normalization is once carried out to described image and is calculated.
Further, the step S1 is specifically included, by original video sequence current frame image f (i, j, t) and former frame Image f (i, j, t-1) carries out optical flow computation and obtains X-direction speed { u (x, y), (x, y) ∈ I } being represented with U, Y-direction speed { v (x, y), (x, y) ∈ I } represented with V, the direction of separating rate obtains four features and is respectively：U to the right+, U to the left-, downward V+With Upward V-, and Gaussian smoothing is carried out to all features：
Above-mentioned formula is carried out and normalized：
The Optical-flow Feature obtained from video sequence carries out similarity S (i, j) calculating：
Wherein,WithPixel pair in two groups of Optical-flow Feature sequences of similarity to be compared is represented respectively Feature calculation result is answered, c is representedWithFour direction after Gaussian smoothing and normalization, i, j is respectively right Answer the frame number in sequence；
The convolution kernel of unit matrix is carried out to S (i, j), is obtained：
Further, the step S2 includes,
S21, video sequence is carried out multistage down-sampled, the L video sequence pyramid of level is formed, to given resolution ratio It is the initialization video sequence f of M × N0(i, j, t), each layer of fl(i, j, t) below equation recurrence calculation：
Wherein, fl(i, j, t) represents per two field picture f (i, j) that in pyramid level l (0≤l≤L) t frames r (m, n) is Gaussian filter,
Wherein,
R (m, n)=r (m) r (n)
R (0)=a, r (1)=r (- 1)=1/4, r (2)=r (- 2)=1/4-2/a；
S22, since the minimum level of resolution ratio, calculate l layers of motion feature sequence fl, hierarchically calculate test specimens This similarity with each sample in training set, as a result chooses k candidate by k nearest neighbor；Then f is calculated at l-1 layersl-1(i, j, t) is right The similarity between motion feature and the previous k candidate selected in l layers is answered, the result of k nearest neighbor is higher for resolution ratio L-2 layers of comparing, untill l=0 layers of resolution ratio highest.
Further, be stored with abnormal driving behavioural characteristic in the property data base for prestoring.
Compared with prior art, beneficial effects of the present invention
Method for detecting fatigue driving based on gaussian pyramid feature of the invention uses gaussian pyramid signature analysis side Method, algorithm complex is low compared to existing technology, computational efficiency is high, recognition accuracy is high.
Brief description of the drawings
Fig. 1 show the method for detecting fatigue driving flow chart based on gaussian pyramid feature of the invention.
Fig. 2 show the testing result diagram in a specific embodiment.
Specific embodiment
With reference to specific embodiment, the present invention is described in further detail.But this should not be interpreted as the present invention The scope of above-mentioned theme is only limitted to following embodiment, and all technologies realized based on present invention belong to model of the invention Enclose.
Embodiment 1：
A kind of method for detecting fatigue driving based on gaussian pyramid feature, comprises the following steps：
S1, the driver's driving image to shooting are pre-processed；
S2, the pretreated view data is carried out down-sampled, obtain multi-resolution Gaussian pyramid feature；
S3, the feature is carried out to be matched with the property data base for prestoring, judge whether driver is fatigue Drive.
Method for detecting fatigue driving based on gaussian pyramid feature of the invention uses gaussian pyramid signature analysis side Method, algorithm complex is low compared to existing technology, computational efficiency is high, recognition accuracy is high.
In a detailed embodiment, the step S1 includes, described image is once carried out light stream, Gaussian smoothing, Normalization is calculated.
In a detailed embodiment, the step S1 is specifically included, by original video sequence current frame image f (i, J, t) and previous frame image f (i, j, t-1) carry out optical flow computation and obtain X-direction speed { u (x, y), (x, y) ∈ I } being represented with U, Y Direction speed { v (x, y), (x, y) ∈ I } represents that the direction of separating rate obtains four features and is respectively with V：U to the right+, to the left U-, downward V+With upward V-, and Gaussian smoothing is carried out to all features：
Above-mentioned formula is carried out and normalized：
The Optical-flow Feature obtained from video sequence carries out similarity S (i, j) calculating：
Wherein,WithPixel pair in two groups of Optical-flow Feature sequences of similarity to be compared is represented respectively Feature calculation result is answered, c is representedWithFour direction after Gaussian smoothing and normalization, i, j is respectively right Answer the frame number in sequence；
The convolution kernel of unit matrix is carried out to S (i, j), is obtained：
In a detailed embodiment, the step S2 includes,
S21, video sequence is carried out multistage down-sampled, the L video sequence pyramid of level is formed, to given resolution ratio It is the initialization video sequence f of M × N0(i, j, t), each layer of fl(i, j, t) below equation recurrence calculation：
Wherein, fl(i, j, t) represents per two field picture f (i, j) that in pyramid level l (0≤l≤L) t frames r (m, n) is Gaussian filter,
Wherein,
R (m, n)=r (m) r (n)
R (0)=a, r (1)=r (- 1)=1/4, r (2)=r (- 2)=1/4-2/a；
S22, since the minimum level of resolution ratio, calculate l layers of motion feature sequence fl, hierarchically calculate test specimens This similarity with each sample in training set, as a result chooses k candidate by k nearest neighbor；Then f is calculated at l-1 layersl-1(i, j, t) is right The similarity between motion feature and the previous k candidate selected in l layers is answered, the result of k nearest neighbor is higher for resolution ratio L-2 layers of comparing, untill l=0 layers of resolution ratio highest.
For Real time identification driver behavior, pair action unrelated with driving is identified and alerts, in order to overcome action of driving The intrinsic difficult point of identification, more accurately recognizes abnormal driving behavior, it is proposed that a kind of motion characteristic pyramid based on light stream is special Levy, and realize the quick acting constrained by the DTW (Dynamic Time Warping) of coarse-to-fine and recognize calculation Method：Carry out light stream, Gaussian smoothing and normalization successively to image first to calculate, be then target to improve overall recognition speed, The light stream sequence data that action video sequence is produced is carried out repeatedly down-sampled and multi-resolution Gaussian pyramid feature is formed, most The Classification and Identification for being acted on the basis of multilayer pyramid feature afterwards.From being identified as action Study of recognition frequently-used data storehouse Basis, discusses the general frame of motion characteristic extraction, action recognition and algorithm, is finally applied to the reality of driver behavior identification In problem, the judgement of abnormal driving behavior is carried out.DTW algorithms proposed by the present invention can in the calculation without opening that consideration is acted Show and two sequences of autoregistration and reduced registering path, more accurately testing result is obtained under requirement of real-time. Method is as follows：A paths are found in similar matrix so that the point on path is all by frame meter most like between two actions Calculate, in path calculation process, preserve each point coordinates and upper some directions, a DTW is calculated after completing, returned by former road Return and remove borderline point and really matched path.
The inventive method makes the speed of Similarity Measure greatly improve compared to existing technology, while because of pyramid adjacent two layers Characteristic sequence similarity is higher, and method also maintains discrimination will not largely be declined.The k for choosing hierachy number and k nearest neighbor is worth working as, Can ensure discrimination be basically unchanged in the case of amount of calculation be greatly reduced, realize rationally efficient being carried out to driving abnormal operation Identification；Specifically, sequence beginning and end is not easy to determine during algorithm of the invention solves the problems, such as traditional DTW methods, energy Enough automatic assorting in calculating process act starting point and end point, so as to improve discrimination, meanwhile, the present invention is special to light stream The preprocess method levied also reduces influence of the noise to recognition result, improves discrimination.
In a detailed embodiment, the abnormal driving behavior that is stored with the property data base for prestoring is special Levy.
Common bad steering behavior, including：(1) watch non-travel direction attentively for a long time, such as watch both sides scene outside window attentively； (2) gear lever is checked when shifting gears；(3) answering cell phone；(4) long-time operation electronic equipment or long-time check instrument etc..
One embodiment of the present of invention provides a kind of sample, and recognition detection has been carried out to following seven kinds of bad steering behaviors, Including：Immerse oneself in be seen after lost article found, rotary head, see left-hand mirror, take a fancy to rearview mirror, see right rear view mirror, catcher machine, bow and see gear.
In advance Sample Storehouse training is carried out to above-mentioned seven kinds of behaviors, and the sample that will be trained is stored, form characteristic According to storehouse.Next using algorithm of the invention to the behavior Similarity Measure in actual driving, the maximum correspondence of selection similarity Behavior corresponding to sequence is used as its recognition result.Fig. 2 show the knot obtained after the test of 126 action sequences of this example Really, wherein completely correct identification 120 is acted, accounting 95.2%, wherein, 6 times wrong identification is respectively：Lost article found is immersed oneself in, is known Left-hand mirror Wei not seen；Seen after rotary head, be identified as taking a fancy to rearview mirror；Bow and see gear, be identified as seeing right rear view mirror；See the right side Rearview mirror, sees after being identified as rotary head；Catcher machine, is identified as seeing right rear view mirror；Seen after rotary head, be identified as seeing right backsight Mirror.Above-mentioned malfunction is analyzed, the fore-aft motion of wrong identification is all mutually related, when movement range is too small, Its Optical-flow Feature relatively, can just be erroneously identified, therefore the Reliability ratio of the method for the present invention is higher, and above-mentioned algorithm is one Can realize completing identification in 3 seconds on the well matched CPU (the present embodiment uses the 2.5GHZ double-core CPU of internal memory 2G) for putting.
After abnormal driving action is identified, i.e., voice reminder is carried out to driver, remind it to stop dangerous driving and act.
Specific embodiment of the invention has been described in detail above in conjunction with accompanying drawing, but the present invention is not restricted to Implementation method is stated, in the case of the spirit and scope for not departing from claims hereof, those skilled in the art can make Go out various modifications or remodeling.

Claims (5)

1. a kind of method for detecting fatigue driving based on gaussian pyramid feature, it is characterised in that comprise the following steps：
S1, the driver's driving image to shooting are pre-processed；
S2, the pretreated view data is carried out down-sampled, obtain multi-resolution Gaussian pyramid feature；
S3, the feature is carried out to be matched with the property data base for prestoring, judge whether driver is fatigue driving.
2. the method for detecting fatigue driving based on gaussian pyramid feature according to claim 1, it is characterised in that described Step S1 includes, light stream, Gaussian smoothing is once carried out to described image, calculating is normalized.
3. the method for detecting fatigue driving based on gaussian pyramid feature according to claim 2, it is characterised in that described Step S1 is specifically included, and original video sequence current frame image f (i, j, t) and previous frame image f (i, j, t-1) are carried out into light stream It is calculated X-direction speed { u (x, y), (x, y) ∈ I } to be represented with U, Y-direction speed { v (x, y), (x, y) ∈ I } is represented with V, The direction of separating rate obtains four features and is respectively：U to the right+, U to the left-, downward V+With upward V-, and all features are carried out Gaussian smoothing：
U ^ + ( x , y ) = U + ( x , y ) ⊗ g ( i , j , 0 , δ ) U ^ - ( x , y ) = U - ( x , y ) ⊗ g ( i , j , 0 , δ ) V ^ + ( x , y ) = V + ( x , y ) ⊗ g ( i , j , 0 , δ ) V ^ - ( x , y ) = V - ( x , y ) ⊗ g ( i , j , 0 , δ ) ,
Above-mentioned formula is carried out and normalized：
U ^ ^ + = U ^ + / | U ^ + | , U ^ ^ - = U ^ - / | U ^ - | V ^ ^ + = V ^ + / | V ^ + | , V ^ ^ - = V ^ - / | V ^ - | ,
The Optical-flow Feature obtained from video sequence carries out similarity S (i, j) calculating：
S ( i , j ) = Σ c = 1 4 Σ x , y ∈ I a c i ( x , y ) · b c j ( x , y )
Wherein,WithPixel correspondence is special in representing two groups of Optical-flow Feature sequences of similarity to be compared respectively Result of calculation is levied, c is representedWithFour direction after Gaussian smoothing and normalization, i, j is respectively correspondence sequence Frame number in row；
The convolution kernel of unit matrix is carried out to S (i, j), is obtained：
S T ( i , j ) = S ( i , j ) ⊗ I ( T ) .
4. the method for detecting fatigue driving based on gaussian pyramid feature according to claim 1, it is characterised in that described Step S2 includes,
S21, video sequence is carried out multistage down-sampled, form the L video sequence pyramid of level, be M to given resolution ratio The initialization video sequence f of × N0(i, j, t), each layer of fl(i, j, t) below equation recurrence calculation：
f l ( i , j , t ) = Σ m = - 2 2 Σ n = - 2 2 r ( m , n ) f l - 1 ( 2 i + m , 2 j + n , t ) ,
Wherein, fl(i, j, t) represents that, per two field picture f (i, j) in pyramid level l (0≤l≤L) t frames, r (m, n) is filtered for Gauss Ripple device,
Wherein,
R (m, n)=r (m) r (n)
R (0)=a, r (1)=r (- 1)=1/4, r (2)=r (- 2)=1/4-2/a；
S22, since the minimum level of resolution ratio, calculate l layers of motion feature sequence fl, hierarchically calculate test sample and The similarity of each sample in training set, as a result chooses k candidate by k nearest neighbor；Then f is calculated at l-1 layersl-1(i, j, t) is to meeting the tendency of Similarity between dynamic feature and the k candidate for previously having been selected in l layers, the result of k nearest neighbor is used for higher l-2 layers of resolution ratio Comparing, untill l=0 layers of resolution ratio highest.
5. the method for detecting fatigue driving based on gaussian pyramid feature according to claim 1, it is characterised in that described Be stored with abnormal driving behavioural characteristic in the property data base for prestoring.
CN201611062106.XA 2016-11-24 2016-11-24 A kind of method for detecting fatigue driving based on gaussian pyramid feature Pending CN106778528A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201611062106.XA CN106778528A (en) 2016-11-24 2016-11-24 A kind of method for detecting fatigue driving based on gaussian pyramid feature

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201611062106.XA CN106778528A (en) 2016-11-24 2016-11-24 A kind of method for detecting fatigue driving based on gaussian pyramid feature

Publications (1)

Publication Number Publication Date
CN106778528A true CN106778528A (en) 2017-05-31

Family

ID=58910961

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201611062106.XA Pending CN106778528A (en) 2016-11-24 2016-11-24 A kind of method for detecting fatigue driving based on gaussian pyramid feature

Country Status (1)

Country Link
CN (1) CN106778528A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108334876A (en) * 2018-05-09 2018-07-27 华南理工大学 Tired expression recognition method based on image pyramid local binary pattern
CN110543848A (en) * 2019-08-29 2019-12-06 交控科技股份有限公司 Driver action recognition method and device based on three-dimensional convolutional neural network
CN110718067A (en) * 2019-09-23 2020-01-21 浙江大华技术股份有限公司 Violation behavior warning method and related device

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101770568A (en) * 2008-12-31 2010-07-07 南京理工大学 Target automatically recognizing and tracking method based on affine invariant point and optical flow calculation
CN103489010A (en) * 2013-09-25 2014-01-01 吉林大学 Fatigue driving detecting method based on driving behaviors
CN103514448A (en) * 2013-10-24 2014-01-15 北京国基科技股份有限公司 Method and system for navicular identification
US20140226913A1 (en) * 2009-01-14 2014-08-14 A9.Com, Inc. Method and system for matching an image using image patches
CN104331151A (en) * 2014-10-11 2015-02-04 中国传媒大学 Optical flow-based gesture motion direction recognition method

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101770568A (en) * 2008-12-31 2010-07-07 南京理工大学 Target automatically recognizing and tracking method based on affine invariant point and optical flow calculation
US20140226913A1 (en) * 2009-01-14 2014-08-14 A9.Com, Inc. Method and system for matching an image using image patches
CN103489010A (en) * 2013-09-25 2014-01-01 吉林大学 Fatigue driving detecting method based on driving behaviors
CN103514448A (en) * 2013-10-24 2014-01-15 北京国基科技股份有限公司 Method and system for navicular identification
CN104331151A (en) * 2014-10-11 2015-02-04 中国传媒大学 Optical flow-based gesture motion direction recognition method

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
WEIHUA ZHANG ETC: "Action Recognition by Joint Spatial-Temporal Motion Feature", 《JOURNAL OF APPLIED MATHEMATICS》 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108334876A (en) * 2018-05-09 2018-07-27 华南理工大学 Tired expression recognition method based on image pyramid local binary pattern
CN110543848A (en) * 2019-08-29 2019-12-06 交控科技股份有限公司 Driver action recognition method and device based on three-dimensional convolutional neural network
CN110543848B (en) * 2019-08-29 2022-02-15 交控科技股份有限公司 Driver action recognition method and device based on three-dimensional convolutional neural network
CN110718067A (en) * 2019-09-23 2020-01-21 浙江大华技术股份有限公司 Violation behavior warning method and related device

Similar Documents

Publication Publication Date Title
CN109784150B (en) Video driver behavior identification method based on multitasking space-time convolutional neural network
CN105354986B (en) Driver's driving condition supervision system and method
CN110378236B (en) Vehicle identity recognition model construction and recognition method and system based on deep learning
Satzoda et al. Multipart vehicle detection using symmetry-derived analysis and active learning
Bertozzi et al. Pedestrian detection for driver assistance using multiresolution infrared vision
Gavrila et al. Vision-based pedestrian detection: The protector system
CN109711316A (en) A kind of pedestrian recognition methods, device, equipment and storage medium again
CN109815867A (en) A kind of crowd density estimation and people flow rate statistical method
CN110276253A (en) A kind of fuzzy literal detection recognition method based on deep learning
EP2395478A1 (en) Monocular 3D pose estimation and tracking by detection
CN108898078A (en) A kind of traffic sign real-time detection recognition methods of multiple dimensioned deconvolution neural network
CN110097044A (en) Stage car plate detection recognition methods based on deep learning
CN106651913A (en) Target tracking method based on correlation filtering and color histogram statistics and ADAS (Advanced Driving Assistance System)
CN109190513A (en) In conjunction with the vehicle of saliency detection and neural network again recognition methods and system
CN110210474A (en) Object detection method and device, equipment and storage medium
CN106778528A (en) A kind of method for detecting fatigue driving based on gaussian pyramid feature
CN107891808A (en) Driving based reminding method, device and vehicle
CN110197152A (en) A kind of road target recognition methods for automated driving system
CN108288047A (en) A kind of pedestrian/vehicle checking method
CN106934355A (en) In-car hand detection method based on depth convolutional neural networks
Rajaram et al. Looking at pedestrians at different scales: A multiresolution approach and evaluations
CN109624918A (en) A kind of safety belt is not system for prompting and method
CN114550270A (en) Micro-expression identification method based on double-attention machine system
CN115294548B (en) Lane line detection method based on position selection and classification method in row direction
US20220019776A1 (en) Methods and systems to predict activity in a sequence of images

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20170531