CN106650635A - Method and system for detecting rearview mirror viewing behavior of driver - Google Patents

Abstract

The invention discloses a method and a system for detecting a rearview mirror viewing behavior of a driver. The method comprises the steps of acquiring a cab image by a vehicle-mounted video system; taking a face and neck visible skin region of the driver as an image analysis processing object; setting an algorithm for adaptive learning and tracing identification of the face and neck visible skin region; extracting a face and neck visible skin outer contour; taking a lowest point of the neck of the face and neck visible skin outer contour as a basic point; dividing the face and neck visible skin outer contour into a left part and a right part by a vertical line passing the basic point; defining an area ratio of the left and right parts as a feature parameter; obtaining a reference eigenvalue of the feature parameter; and judging whether the rearview mirror viewing behavior exists or not according to a numerical relationship between the current feature parameter and the reference eigenvalue in combination with a stare threshold and a duration threshold. The detection system is simple in structure; and the detection method has relatively high accuracy rate and good adaptive learning capability and anti-jamming capability, is high in detection efficiency, and can meet real-time detection and reminding requirements of the rearview mirror viewing behavior during quick running of a vehicle.

Description

A kind of driver's rearview mirror checks behavioral value method and system

Technical field

The present invention relates to technical field of vehicle safety, can be in turn inside diameter, doubling, change more particularly, to one kind During the steering manipulations such as track, whether automatic detection driver has effectively checked rearview mirror, and makes the side of corresponding prompting accordingly Method and system.

Background technology

As increasing sharply for vehicle guaranteeding organic quantity, vehicle accident incidence rate remain high, traffic safety has become universal The problem being concerned.Statistical data shows that 25%～30% vehicle accident is directly related with the state of attention of driver, its The steering manipulation process such as middle turn inside diameter, doubling, change lane is one of main generation occasion of vehicle accident, is especially common in Driver does not note the situation of Vehicular turn side rear transport information.Rearview mirror checks that the real-time detection of behavior and necessary prompting are The path explored of such vehicle accident probability of happening is reduced, its research is occupied for the mixed traffic environment with China as representative Especially there is for the developing country of many, traffic participant type diversification realistic meaning.

At present, vehicle-mounted vision and image processing techniquess are the main realities that current driver's rearview mirror checks behavioral value research Existing scheme, its advantage are deliberately to coordinate and be difficult to disturb driver without the need for driver.In existing research and technology application, drive The identification of member's head pose is key therein, is based primarily upon the side such as face template, facial detail feature and face mask feature Method is estimating attitude parameter.Due to applying faceform and/or minutia, algorithm is relative complex to be easily caused vehicle high-speed row Detect when sailing that real-time is not enough.Although detailed information ability to express is reduced only for the method for face mask, make algorithm real-time Property improved, but face geometric properties when defining the selection of positioning datum tend to be disturbed.Additionally, these situations are certainly It has been short of in terms of adaptive learning ability, has been required for greatly relying on the template being closely related with each driver or the feature for pre-setting Parameter, but when driver's shape of face change (as change driver when) facial detail change greatly (when blocking such as adornment) or into As during Angulation changes, (during such as adjustment seat) Detection accuracy is possible impacted.

CN201110394497.6 discloses a kind of " safety driving monitoring system and method based on pattern recognition ", the party Method is referred to and the face of driver is positioned and split by blee detection method, but finally or in order to carry out face The extraction of feature, the defect such as equally exists poor real, is easily disturbed.

As can be seen here, study a kind of driving with adaptive learning ability, real-time and good capacity of resisting disturbance The person's of sailing rearview mirror checks that behavioral value is necessary with based reminding method.

The content of the invention

The technical problem to be solved is to provide a kind of driver's rearview mirror to check behavioral value method and be System, which selects special image analysis processing object, using little datum mark is changed, can reach adaptive learning ability it is strong, The purpose of real-time and strong antijamming capability.

To solve above-mentioned technical problem, the technical solution of the present invention is：

A kind of driver's rearview mirror checks behavioral value method and system, and with vehicle carrying video system, T is real-time at timed intervals It is continuous to gather driver's cabin image frame by frame, with the visible skin areas of driver face and cervical region collectively as image analysis processing pair As setup algorithm method comes adaptive learning and Tracking Recognition driver's face neck visible skin areas, and extracting driver's face neck can See skin outline, as basic point, the vertical line for crossing the basic point can by face neck for the cervical region minimum point with face neck visible dermis outline See that skin outline is divided into left and right two parts, define left and right two parts area ratio and be characterized parameter, obtain this characteristic parameter Reference characteristic value, by the numerical relation of current signature parameter and reference characteristic value and with reference to staring threshold value and duration threshold To determine whether that there is rearview mirror checks behavior.

Preferably, the initial face of driver is recognized using frames differencing searching algorithm adaptive learning during vehicle launch Neck visible skin areas.

Preferably, the driver's face neck visible skin areas based on prior image frame during vehicle traveling are known using breathing method Current face neck visible skin areas are not tracked.

Preferably, the cumulative probability local peaking distribution rule checked in action process in rearview mirror using the characteristic parameter Restrain to estimate reference characteristic value.

A kind of driver's rearview mirror checks behavioral value method, specifically includes following steps：

Step one：When a vehicle is started, can using the initial face neck of frames differencing searching algorithm self-adapting estimation driver See skin area and calculate face neck skin initial gray average in the region；

Step 2：Vehicle travel during, driver's cabin video system whenever sample a frame new images, based on prior image frame Driver's face neck visible skin areas locating rectangle quickly recognizes current face neck visible skin areas using breathing method, while completing Current face neck skin gray average estimation；If now vehicle is not turned to, system is in face neck region dynamic tracking state, and Repeat step two；Otherwise continue step 3；

Step 3：In the face neck identification positioning region internal calculation contour feature parameter of present image；

Step 4：Local peaking is calculated according to the cumulative probability of face neck visible dermis outline characteristic parameter and estimates profile base Quasi- eigenvalue；

Step 5：Calculated according to the numerical relation of current outline characteristic parameter and reference characteristic value and stare coefficient, according to solidifying The persistent period that driver observes rearview mirror state is counted depending on threshold condition, is further driven according to duration threshold condition criterion Whether the person of sailing implements effective rearview mirror and checks behavior.

Preferably, in the presence of judging that the effective rearview mirror of driver checks behavior, the institute during being checked using this There is characteristic parameter to update cumulative distribution probability.

Preferably, the initial face neck visible skin areas recognition methodss of driver are as follows：

The first step：Frame difference source images are searched for and do difference；Two image P are searched for i.e. in initial learn image₀₁And P₀₂, Driver head's attitude of this two images has significant difference and hand gestures difference is not obvious, and the two field pictures are made the difference Point, obtain frame difference result；

Second step：Frame difference result binaryzation；Note binaryzation result images are B₀, i.e.,：

Wherein b₀(i, j) represents image B₀The i-th row jth row pixel, ξ is binary-state threshold；

3rd step：Tiny speckle and gossamer are removed；Image B is removed₀Tiny speckle and gossamer, obtain image A₀；

4th step：Face neck visible skin areas are positioned；Calculate image A₀Boundary rectangle R₀₁, which will in frame difference source images Driver face is surrounded, by rectangle R₀₁Certain multiple is extended downwardly highly for rectangle R₀₂, R₀₂To surround simultaneously driver face and Cervical region, defines R₀₂For driver's face neck area-of-interest, that is, obtain described face neck visible skin areas；

5th step：Face neck skin gray average is calculated；Calculate area-of-interest R₀₂ε span grey level histograms, that is, distinguish Statistics gray value is located at interval pixel ratio, corresponds to the hair gray scale peak value of driver in rectangular histogram with two peak values respectively Or face neck skin gray scale peak value, skin gray scale peak value g is selected according to practical situation₀, and calculate [g₀-ε,g₀+ ε] between pixel ash Degree average can obtain face neck skin gray average

6th step：Initial face neck visible skin areas identification positioning；According toIn R₀₂It is bilateral threshold filtering two in inside Value, can obtain the face neck pachylosiies image H of driver₀′：

Wherein：h′₀(i, j) represents image H '₀The i-th row jth row pixel,H′₀Weight Multiple three step process, filtering interfering obtain face neck skin image H₀, its boundary rectangle R₀As frame difference source images P₀₁In face neck Visible skin areas, i.e., initial face neck visible skin areas.

Preferably, the concrete search procedure of frame difference source images is：

Any two field pictures P that acquired image during vehicle launch is concentrated_uAnd P_vDifference is done, note frame difference result is D：

D (i, j)=| P_u(i,j)-P_v(i,j)| (1)

In formula (1), i and j is the row and column of pixel respectively；D is horizontally divided into into D₁And D₂Upper and lower two parts, the ration of division is γ；Calculate D₁And D₂Pixel grey scale accumulation compare ω：

In formula (2), d (i, j) represents the pixel of the i-th row jth row of image D, and W and H is respectively the width and height of image D Degree；

Given coefficient of determination Ω, the D if ω >=Ω₁The time gradient information for containing is noticeably greater than D₂, i.e. head pose change Change and be noticeably greater than hand gestures change, by P_uAnd P_vAs frame difference source images P₀₁And P₀₂, search completes；Otherwise continue random frame poor Search；If being arbitrarily still unsatisfactory for ω >=Ω after two field pictures difference, two field pictures when taking ω maximums are frame difference source figure Picture；Now frames differencing result is D to note₀, D₀=| P₀₁-P₀₂|。

Preferably, the breathing recognition methodss of driver's face neck visible skin areas domain are as follows：

A two field picture is gathered every T time, note kth time sampled images are P_k, driver's face neck region be R_k, face neck skin Gray average isK >=1, P₀=P₀₁For initial condition, initial parameter R₀WithAsked according to frame difference algorithm in vehicle launch ；

The first step：Face neck area-of-interest is positioned；With R_k-1Center (x_k-1,y_k-1) for basic point, to R_k-1Carry out expansion expansion , obtain present image P_kIn driver face neck area-of-interest R_k', makeThe expansion of note width and height The factor is respectively α and β, then：

w_k=w_k-1+Round(α·T)+2 (5)

h_k=h_k-1+Round(β·T)+2 (6)

Wherein w_k-1And h_k-1Respectively region R_k-1Width and height, w_kAnd h_kRespectively region R '_kWidth and height, Round (.) is bracket function；

Second step：Face neck skin gray average is calculated；Due to the face neck skin gray scale in image in little sampling time interval T Change in Mean is by very little, therefore R '_kIt is located in the ε span grey level histograms in region Interval peak value g_kMust For the face neck skin gray scale peak value of present image, face neck skin gray averageAs [g_k-ε,g_k+ ε] interval pixel grey scale Average；

3rd step：Face neck visible skin areas are recognized；In face neck area-of-interest R '_kInside, does bilateral threshold filtering simultaneously Binaryzation, extracts the coarse image H ' of driver's face neck visible dermis_k：

Wherein h '_k(i, j) represents image H '_kThe i-th row jth row pixel,To H '_k Carry out filtering fleck and gossamer interference, the final binary image H for obtaining face neck visible dermis_k, its boundary rectangle R_kAs Image P_kIn face neck visible skin areas.

Preferably, tiny speckle and the concrete removing method of gossamer are：

First to image B₀Morphological image opening operation is carried out, is removed on the premise of image area is not substantially changed tiny Speckle and the connection in disconnection gossamer and other regions, note result are image C₀：

C₀=Open (B₀,E) (11)

In formula (11), E is 3 × 3 rectangle kernels that reference point is located at center；

Secondly calculate image C₀The area girth ratio in middle non-interconnected region, filters ratio less than area girth than threshold value r's All gossamers, obtain image A₀；

To image H '₀With H '_kWhen being processed, by the B in formula (11)₀Replace with H '₀With H '₀。

Preferably, face neck visible dermis outline calculation of characteristic parameters and reference characteristic value evaluation method are as follows：

The closed curve of face neck visible dermis outline is obtained in the face neck visible skin areas of gained, first by cervical region Reference mark O of the low spot as contour line_k；

Secondly, cross O_kVertical line face neck outline is divided into into left and right two parts, its area S_kLAnd S_kRIt is relevant with head pose, Area than φ is：

φ=S_kR/S_kL (8)

φ_L、φ₀And φ_RLeft and right area ratio when observation left-hand mirror, dead ahead and right rear view mirror, definition are corresponded to respectively Eigenvalue on the basis of three；

By T sampled images interval time, parameter phi and its tire out that continuous l rearview mirror effectively checked in action process are calculated Product probability, obtains φ using the local peaking that cumulative probability is distributed_L、φ₀And φ_REstimated value.

Preferably, rearview mirror checks that behavior asset pricing decision method is：

Definition respectively stares coefficient when checking left and right rearview mirror：

μ_L=| φ-φ_L|/φ_L (9)

μ_R=| φ-φ_R|/φ_R (10)

Work as μ_L≤μ_L0When confirm that driver eye movement fixation point is located at vehicle left-hand mirror, work as μ_R≤μ_R0When confirm driver's eye Dynamic fixation point is located at vehicle right rear view mirror, wherein μ_L0And μ_R0Threshold value is stared respectively；Statistics meets μ respectively_L≤μ_L0And μ_R≤μ_R0 The number n of the continuous sampling image of condition_LAnd n_R, and if only if stares duration T n of left-hand mirror_L≥T_L0When judge drive The person of sailing observes the effectiveness of left-hand mirror, and and if only if stares duration T n of right rear view mirror_R≥T_R0When judge driver The effectiveness of observation right rear view mirror, wherein T_L0And T_R0Respectively duration threshold.

A kind of detecting system according to above-mentioned detection method, including system switching, power module, image capture module and control Device processed；The system switching is used to be turned on and off system；The power module is used for as system and each module for power supply；The image acquisition Module is for being acquired to driving indoor driver's image；The controller plc is to collect to image capture module Image is processed, and judges whether driver effectively checks rearview mirror.

A kind of driver's rearview mirror checks behavioral value system, including system switching, power module, image capture module and Controller；The system switching is used to be turned on and off system；The power module is used for as system and each module for power supply；The image is adopted Collect module for being acquired to driving indoor driver's image；The controller plc is to collect to image capture module Image processed, and judge whether driver effectively checks rearview mirror, which includes frame difference evaluator, breathing evaluator, face Neck contour feature parameter calculator, profile reference characteristic value estimator and rearview mirror check behavior asset pricing determinant；The frame difference is known Other device completes gray average study and initial face neck of driver's face neck skin under the conditions of current light source in vehicle launch Visible skin areas adaptable search is recognized；It is equal that the breathing evaluator completes driver's face neck skin gray scale during vehicle traveling Value study and the quick Tracking Recognition of face neck visible skin areas；The face neck contour feature parameter calculator is used to calculate current collection Image in driver face neck outline characteristic parameter；The profile reference characteristic value estimator is using the tired of the characteristic parameter Accumulate probability to estimate profile reference characteristic value；The rearview mirror checks face neck outline of the behavior asset pricing determinant according to present image Characteristic parameter and profile reference characteristic value judge whether driver implements effective rearview mirror and check behavior.

Preferably, described controller further includes cumulative probability renovator, and the cumulative probability renovator is used for real-time Update the accumulation probability of occurrence of face neck contour feature parameter.

Preferably, described controller further includes PORT COM, and the PORT COM receives vehicle other Electronic Controls The Vehicular turn information that unit transmission comes, while will judge that conclusion is sent to other side and further uses.

After such scheme, the present invention is compared with traditional method, and which is mainly applied when using driver's physical trait Be driver face and cervical region outline curve, it is special different from existing research and the utilization face template in invention, details Face mask of seeking peace line, it is easier to distinguish with outside other regions, extracts more convenient, and fast response time, real-time are higher；And Not labile cervical region minimum point is adopted for basic point, the face neck outline left and right area ratio of basic point vertical line division is crossed as spy Parameter is levied, therefore capacity of resisting disturbance is higher.

Additionally, frame difference method adaptive learning when the present invention is divided into vehicle launch in terms of the identification positioning of face neck region and The quick Tracking Recognition of breathing method when vehicle is travelled；It is the cumulative probability using contour feature parameter when reference characteristic value is given The regularity of distribution is calculating renewal in real time；Check that behavior is while basis stares threshold value and duration threshold when judging in rearview mirror Come carry out.

In a word, present configuration is simple, and detection method has higher accuracy rate, good adaptive learning ability and resists Interference performance, and detect that real-time is high, the rearview mirror during disclosure satisfy that vehicle fast running check the real-time detection of behavior and Remind and require.

Description of the drawings

Fig. 1 is the composition schematic diagram of system of the present invention；

Fig. 2 is the flow chart that driver's rearview mirror of the present invention checks behavioral value method；

Fig. 3 is step one face neck region frame difference recognition result schematic diagram in detection method of the present invention；

Fig. 4 is step 2 face neck region breathing recognition result schematic diagram in detection method of the present invention；

Fig. 5 is step 3 face neck region outline curve synoptic diagram in detection method of the present invention；

Fig. 6 is the typical change rule of face neck outline when step 3 checks right rear view mirror in detection method of the present invention Schematic diagram；

Fig. 7 is the typical change schematic diagram of parameter phi when step 3 checks rearview mirror in detection method of the present invention；

When Fig. 8 is step 4 l=8 in detection method of the present invention, the probability distribution of the parameter phi of driving video is illustrated Figure；

The change curve of each reference characteristic value when Fig. 9 is step 4 l=8 in detection method of the present invention.

Specific embodiment

The invention will be further described with specific embodiment below in conjunction with the accompanying drawings.

Disclosed is a kind of driver's rearview mirror checks behavioral value system, as shown in figure 1, for institute of the present invention State the preferred embodiment of system.The detecting system includes system switching, power module, image capture module, display screen, ginseng Number setup module and controller.Wherein：

Described system switching is used to be turned on and off system, may preferably be normally closed type switch, i.e. vehicle power supply and starts System default work afterwards.

Described power module may preferably be Vehicular accumulator cell for being system and each module for power supply.

Described image capture module for being acquired to driving indoor driver's image, its can include photographic head, Light source and fixing device for installing.Described light source may preferably be the infrared light supply that people are difficult to perceive, the suitable shooting Head may preferably be infrared camera, it is possible to be installed in front of driver or shelves glass before the vehicle of diagonally forward, it is desirable to can Photograph complete driver's cabin image.

Described display screen is used for display system operational factor and video flowing, and can provide good people to parameter setting Machine information interacts channel.

For arranging driver's essential information, characteristic information and rearview mirror, described parameter setting module checks that behavior is sentenced All kinds of Image Processing parameters being related in determining algorithm and threshold value.

Described controller is to process to the image that image capture module is collected, and judges whether driver has Effect checks rearview mirror.Which can include frame difference evaluator, breathing evaluator, face neck contour feature parameter calculator, profile benchmark Eigenvalue estimator and rearview mirror check behavior asset pricing determinant, can also further include that cumulative probability renovator and communication connect Mouthful.Described frame difference evaluator completes gray average of driver's face neck skin under the conditions of current light source in vehicle launch Practise and initial face neck region adaptable search is recognized, be that the continuous quick face neck region Dynamic Recognition during vehicle is travelled is established Parameter basis.Described breathing evaluator completes the study of driver's face neck skin gray average and face neck region during vehicle traveling The quick Tracking Recognition in domain, is that the extraction of face neck visible dermis profile and feature analysiss establish image parameter basis.Described face neck Contour feature parameter calculator is used for the face neck outline characteristic parameter of driver in the image for calculate current collection.Described wheel Driver face neck outline of the wide reference characteristic value estimator using the calculated acquisition of the face neck contour feature parameter calculator The cumulative probability of characteristic parameter is estimating profile reference characteristic value.Described rearview mirror checks behavior asset pricing determinant according to current The face neck outline characteristic parameter of image and profile reference characteristic value judge whether driver implements effective rearview mirror and check Behavior.Described cumulative probability renovator is used for the accumulation probability of occurrence of real-time update face neck contour feature parameter.Described is logical The next Vehicular turn information of other electronic control units transmission of news port reception vehicle, while will judge that conclusion is sent to other side and enters One step is used.

Invention further discloses a kind of driver's rearview mirror checks behavioral value method：With vehicle carrying video system on time Between be spaced T real-time continuous gather driver's cabin image frame by frame, with the visible skin areas of driver face and cervical region collectively as figure As analyzing and processing object, setting image processing algorithm comes adaptive learning and tracking driver's face neck visible skin areas, carries Driver's face neck visible dermis outline is taken, the cervical region minimum point with face neck visible dermis outline crosses the basic point as basic point Face neck visible dermis outline is divided into left and right two parts by vertical line, is defined left and right two parts area ratio and is characterized parameter, obtains This characteristic parameter reference characteristic value, by the numerical relation of current signature parameter and reference characteristic value and with reference to staring threshold value Determine whether that with duration threshold there is rearview mirror checks behavior.

Concrete steps are as shown in Figure 2：

Step one：When a vehicle is started, can using the initial face neck of frames differencing searching algorithm self-adapting estimation driver See skin area (abbreviation face item region) and calculate face neck skin initial gray average, be that the continuous Dynamic Recognition of face neck region is carried out Prepare.

Step 2：Vehicle travel during, driver's cabin video system whenever sample a frame new images, based on prior image frame Driver's face neck region locating rectangle quickly recognizes current face neck region using breathing method, while completing current face neck skin gray scale Average is estimated.If now vehicle is not turned to (turn signal state transmitted according to communication interface or steering wheel angle determine), System is in face neck region dynamic tracking state, and repeats step 2；Otherwise continue step 3.

Step 3：In the face neck identification positioning region internal calculation contour feature parameter of present image.

Step 4：Local peaking is calculated according to the cumulative probability of face neck visible dermis outline characteristic parameter and estimates profile base Quasi- eigenvalue.

Step 5：Calculated according to the numerical relation of current outline eigenvalue and reference characteristic value and stare coefficient, according to staring Threshold condition statistics driver observes the persistent period of rearview mirror state, is further driven according to duration threshold condition criterion Whether member implements effective rearview mirror is checked behavior.When judging not exist, driver is reminded to note the traffic of turn side rear Situation；Further, in the presence of judgement, all eigenvalues during checking using this update cumulative distribution probability. Return to step two afterwards.

Such iterative cycles, until vehicle stall.

More specifically, in each step of the above：

(1), in the step one, the concrete grammar of the initial face neck region frame difference identification of driver during vehicle launch is as follows：

The first step：Frame difference source images are searched for and do difference.The initial learn image set for being gathered during vehicle launch Two image P of middle search₀₁And P₀₂, driver head's attitude of this two images there is significant difference and hand gestures difference not Substantially, and to the two field pictures difference is done, frame difference result is obtained.

Concrete search procedure is, by any two field pictures P_uAnd P_vDifference is done, note frame difference result is D：

D (i, j)=| P_u(i,j)-P_v(i,j)| (1)

In formula (1), i and j is the row and column of pixel respectively.D is horizontally divided into into D₁And D₂Upper and lower two parts, the ration of division is γ.Calculate D₁And D₂Pixel grey scale accumulation compare ω：

In formula (2), d (i, j) represents the pixel of the i-th row jth row of image D, and W and H is respectively the width and height of image D Degree.

Given coefficient of determination Ω, the D if ω >=Ω₁The time gradient information for containing is noticeably greater than D₂, i.e. head pose change Change and be noticeably greater than hand gestures change, by P_uAnd P_vAs frame difference source images P₀₁And P₀₂, search completes；Otherwise continue random frame poor Search.If being arbitrarily still unsatisfactory for ω >=Ω after two field pictures difference, two field pictures when taking ω maximums are frame difference source figure Picture.Now frames differencing result is D to note₀, D₀=| P₀₁-P₀₂|。

Second step：Frame difference result binaryzation.Note binaryzation result images are B₀, i.e.,：

Wherein b₀(i, j) represents image B₀The i-th row jth row pixel, ξ is binary-state threshold.

3rd step：Tiny speckle and gossamer are removed.Image B is removed₀Tiny speckle and gossamer, obtain image A₀。

First to image B₀Morphological image opening operation is carried out, is removed on the premise of image area is not substantially changed tiny Speckle and the connection in disconnection gossamer and other regions, note result are image C₀：

C₀=Open (B₀,E) (11)

In formula (11), E is 3 × 3 rectangle kernels that reference point is located at center.Secondly calculate image C₀Middle non-interconnected region Area girth ratio, filters ratio less than area girth than all gossamers of threshold value r, obtains image A₀。

4th step：Face neck visible skin areas are positioned.Calculate image A₀Boundary rectangle R₀₁, which is in frame difference source images P₀₁ (or P₀₂) in will surround driver face.Consider the face neck typical proportions of human body, by rectangle R₀₁Highly extending downwardly 1.25 times is Rectangle R₀₂, R₀₂Driver face and cervical region will be surrounded simultaneously, driver's face neck area-of-interest will be defined as, i.e., described in acquisition Face neck visible skin areas.Which is internal mainly comprising driver's hair, face, cervical region, adornment, clothes and other in-car foreign objects The image information of body, and when photographic head is located at driver's cabin front upper place, hair, face and cervical region will be primary picture contents.

5th step：Face neck skin gray average is calculated.Calculate area-of-interest R₀₂ε span grey level histograms, that is, distinguish Statistics gray value be located at it is interval (0, ε -1), (ε, 2 ε -1), (2 ε, 3 ε -1) ..., (n ε, 255) } pixel ratio.Wherein ε's Effect is the local inequality and nuance for reducing driver's face neck skin gray scale in image.Main peak value g in obvious rectangular histogram₀₁ With minor peaks g₀₂The hair gray scale peak value or face neck skin gray scale peak value of driver are corresponded to respectively, are specifically dependent upon the figure of the two As size.For simple problem, the present embodiment only considers that driver is nearly black hair, yellow fair-complexioned situation (other feelings Condition can be preset according to practical situation).Being apparent from now hair gray scale less than skin gray scale, therefore skin gray scale peak value is：

g₀=Max (g₀₁,g₀₂) (12)

Calculate [g₀-ε,g₀+ ε] between pixel grey scale average can obtain face neck skin gray average

6th step：Initial face neck region identification positioning.According toIn R₀₂Bilateral threshold filtering binaryzation are done in inside, can Obtain the face neck pachylosiies image H ' of driver₀：

Wherein：To H '₀Repeat three step process, filtering interfering obtains face neck skin image H₀, its boundary rectangle R₀As frame difference source images P₀₁In face neck region.

As a result illustrate as shown in Figure 3.

In Fig. 3：A () and (b) is the frame difference source images for searching；C () is frame difference result；D () is frame difference result binaryzation Image, wherein also there are some very thin tracks moved and staying because of shoulder and arm in addition to comprising head pose variation track； E () is to filter tiny speckle and the image after gossamer, dotted line frame is its boundary rectangle R₀₁；F () shows R₀₁In frame difference source figure Locating effect as in, wherein solid box are R₀₁Face neck area-of-interest R after extending downwardly₀₂；G () is R₀₂ε=10 span Grey level histogram, its primary and secondary peak value are respectively g₀₁≈ 35 and g₀₂≈ 90, can determine whether face neck skin gray scale peak value g accordingly₀≈ 90, enters One step can try to achieve face neck skin gray averageH () is face neck area-of-interest in [83,103] interval bilateral threshold value Filter and binaryzation result figure；I () is to filter fleck and the face neck skin image after gossamer interference, dotted line frame is external for which Rectangle, represents the face neck positioning region in frame difference source images, and effect is as shown in (j).

(2), in the step 2, face neck region breathing recognition methodss are specific as follows：

Infrared video stream is sampled, and a two field picture is gathered every T time.Note kth time sampled images are P_k, driver Face neck region (i.e. face neck visible skin areas) is R_k, face neck skin gray average bek≥1。P₀=P₀₁For initial condition, Initial parameter R₀WithTried to achieve according to frame difference algorithm in vehicle launch, i.e., completed by described step one.

The first step：Face neck area-of-interest is positioned.With R_k-1Center (x_k-1,y_k-1) for basic point, to R_k-1Carry out expansion expansion , obtain present image P_kIn driver face neck area-of-interest R '_k.Due to adjacent image sampling twice interval time T very Little, the mobility scale of driver's face cervical region position is comparatively small, therefore during less expansion rate ensures that present image Driver's face neck visible skin areas R_kPositioned at R '_kInside, i.e.,Expansion rate is more big more reliable, but interested Region R '_kTo be bigger so that image procossing amount is increased, algorithm real-time is deteriorated.Expansion rate is mainly affected by sampling interval T, When T gets over hour, in neighbouring sample image, the moving range of driver's head and neck is less, and expansion rate can be less.Note width w and height The expansion factor of h is respectively α and β, then：

w_k=w_k-1+Round(α·T)+2 (5)

h_k=h_k-1+Round(β·T)+2 (6)

Wherein w_k-1And h_k-1Respectively region R_k-1Width and height, w_kAnd h_kRespectively region R '_kWidth and height, Round (.) is bracket function.

Second step：Face neck skin gray average is calculated.In view of infrared illumination condition and driver in the little sampling interval T The fact that face neck mobility scale is all relatively small, the face neck skin gray average being apparent from neighbouring sample image change very little. Therefore R '_kIt is located in the ε span grey level histograms in regionInterval peak value g_kIt must be the face of present image Neck skin gray scale peak value.Face neck skin gray averageAs [g_k-ε,g_k+ ε] interval pixel grey scale average.

3rd step：Face neck region is recognized.In face neck area-of-interest R '_kInside, does bilateral threshold filtering binaryzation, carries Take the coarse image H ' of driver's face neck visible dermis_k：

WhereinTo H '_kUtilize (11) formula and area girth than threshold value r filter fleck and Gossamer is disturbed, the final binary image H for obtaining face neck visible dermis_k, its boundary rectangle R_kAs image P_kIn face neck region Domain, i.e., current face neck visible skin areas.

From from image processing region, algorithm above is experienced from R_k-1It is expanded into R '_kAgain by R '_kNarrow down to R_kProcess.

Result citing is as shown in Figure 4.

In Fig. 4：A () is -1 sampled images of kth, wherein dashed rectangle R_k-1For the positioning region of its driver's face neck, in Heart coordinate for (301,261), a width of 230 pixel, a height of 310 pixel, face neck skin gray averageB () is adopted for kth time Sampled images, wherein dotted line frame are R_k-1Face neck area-of-interest R ' after expansion_k, centre coordinate still for (301,261), a width of 240 Pixel, a height of 320 pixel；C () is R '_kε=10 span grey level histogram, there is peak value g [87-10,87+10] is interval_k= 85, calculate the face neck skin gray scale obtained in [85-10,85+10] interval pixel grey scale average as kth sampled images equal ValueD () is for R '_kRegion carries out the visible dermis coarse image of bilateral threshold filtering binaryzation, exist it is some with The approximate very thin shape interference of skin gray scale；E () is the image H after filtering interfering_k, dotted line frame is its boundary rectangle R_k, its center is sat Be designated as (299,254), wide 223 pixel, high 322 pixel；F () shows R_kLocating effect in present image.

(3), in the step 3, the circular of face neck visible dermis outline characteristic parameter is as follows：

Directly enter every trade boundary scan in the face neck visible skin areas of gained, obtain face neck visible dermis outline Closed curve.As a result citing as shown in figure 5, respectively for driver usually, wear sunglasses and camouflage color mask when situation.

Fig. 5 results show：Face neck visible dermis outline is not necessarily smooth or round and smooth closed curve, may be because of hair Send out, the interference such as glasses, mask and different numbers and shape is presented.Therefore profile spy is defined from aspects such as the yardsticks and shape of curve Possible less effective is levied, area is surrounded according to curve Comparatively speaking more feasible come defined feature parameter.

As a example by checking vehicle right rear view mirror, the typical change of face neck visible dermis contour curve is as shown in Figure 6.Wherein： A () is checked in front of vehicle heading for driver；B () is to turn one's head to the right；C () is further to turn one's head to the right；D () is right for observation Rearview mirror state；E () is holding right rear view mirror observation state；F () terminates from right torsion forward to turn one's head for observation；G () is further Turn round forward and turn one's head；H () is checked in front of vehicle heading to return to.As can be seen from the figure：

First, cervical region low spot near zone is in image P_kIn location and shape change all very little, cervical region low spot can be made For reference mark O of contour line_k。O_kOnly there is large variation when driver significantly sways (such as stretch forward and take thing), and this is in car Traveling during belong to the risk behavior that should forbid；When driver's correctly wear safety belt, O_kAmount of change will be further Limited.This exactly the application by driver face and cervical region collectively as analysis object the reason for, cervical region plays key reference Effect, it is different from the thinking in tradition research only for face.

Secondly, cross O_kVertical line face neck outline is divided into into left and right two parts, its area S_kLAnd S_kRIt is relevant with head pose, Area than φ is：

φ=S_kR/S_kL (8)

φ during certain driver or so rearview mirror is continuously checked with head rotational angle theta typical change rule such as Fig. 7 institutes Show.

In figure, θ=0 represents not turning one's head right ahead state of check.Curve MR is turned one's head for driver to the right after checking the right side The process of visor；RM is checked for right rear view mirror and is finished later process；ML is the process turned one's head to the left and check left-hand mirror；LM is a left side Rearview mirror is checked and finishes later process.φ_L、φ₀And φ_RLeft side when observation left-hand mirror, dead ahead and right rear view mirror is corresponded to respectively The factors such as right area ratio, concrete numerical value and driver's face neck, vehicle width, seat position and driving habit are relevant.Even if taking the photograph As head is located at driver Lian Jing centers segmentation plane φ₀1 may not be also equal to.Fig. 7 results show：Check when driver turns one's head to the right During right rear view mirror process, φ values are by φ₀It is incremented to φ_R, during observing right rear view mirror, then it is maintained at φ_RFuctuation within a narrow range near value, looks into See when finishing later process by φ_RIt is decremented to φ₀；And in the case of left-hand mirror is checked, φ value changes are respectively by φ₀Successively decrease To φ_L, be maintained at φ_LNearby fuctuation within a narrow range, by φ_LIt is incremented to φ₀.It can be seen that, parameter phi it is continuous during rearview mirror is checked and With dull increase and decrease characteristic, can be used as a characteristic parameter of face neck visible dermis outline.Define φ_L、φ₀And φ_RFor base Quasi- eigenvalue.According to φ and φ_LAnd φ_RNumerical relation can judge driver whether in checking that left and right backsight is specular respectively State.

(4), in the step 4, the concrete grammar of the cumulative probability local peaking regularity of distribution estimation of reference characteristic value is such as Under：

For different drivers or vehicle, due to face neck difference it is different with photographic head installation site, reference characteristic value φ_LWith φ_RIt is different；Even if same driver drives same vehicle, as the reasons such as hair style, adornment, seat adjustment also lead to φ_LWith φ_RChange.Obviously the given reference characteristic value of the way of traditional preset parameter or template cannot be passed through.

To propose φ_LAnd φ_RAdaptive learning method, with vehicle turn signal open and terminate as the whole story mark citing one The secondary rearview mirror for completing is turned one's head and checks behavior complete procedure, by T sampled images interval time, calculates continuous l rearview mirror effective The face neck profile left and right area checked in action process is than parameter phi and its cumulative probability.By taking l=8 as an example, certain driving video piece The probability distribution curve of the φ of section is as shown in Figure 8.In Fig. 8, abscissa represents the codomain scope of φ, and vertical coordinate represents different φ values The probability of corresponding appearance, the local peaking of L, F, U and R for probability curve, Jing eye trackers confirm actual point of these local peakings Do not represent driver in checking in front of the outer left-hand mirror of car, travel direction, (the application does not analyze this to in-car upper right side rearview mirror Kind of situation) and car right rear view mirror outward state.This is because：Effectively rearview mirror is checked and is necessarily required to persistently coagulating for certain time length Depending on to promote the understanding of driver's offside rear transport information, period will collect multiple it is similar stare image, therefore can obtain Much approximate φ values of individual numerical values recited；Although driver checks that the action of rearview mirror may not be very consistent every time, there is torsion Excessive or very few situation, head also likely to be present weak vibrations, but if the occurrence number of all φ values that add up calculating Probability of occurrence, then necessarily draw probability peak.This characteristic is referred to as into driver's rearview mirror and checks that the face neck outline of behavior is special Levy l cumulative probability local peaking rule of parameter phi.φ can be obtained using these local peakings_L、φ₀And φ_REstimated value, As in Fig. 8, sign is respectively 0.69,0.98 and 1.32.

After rearview mirror observed behavior is obtained to be confirmed, accumulation should be updated using this all φ value during checking in time Probability simultaneously re-starts parameter estimation.As generation S_kLAnd S_kRWhen numerical value is mutated, driver's face neck visible skin areas size is judged There is large change, now delete existing cumulative probability and re-start probability learning.When Fig. 9 is l=8, the driver is in vehicle row Reference characteristic value φ before and after sunglasses is worn during sailing_L、φ₀And φ_RChange curve, wherein abscissa represents that driver is effective The number of times of observation rearview mirror, vertical coordinate represent the φ estimated according to aforementioned cumulative probability local peaking rule_L、φ₀And φ_R.Can To find out：T after the 25th rearview mirror checks behavior₀At the moment, cause each reference characteristic value generation larger due to wearing sunglasses Change, but only need 1 rearview mirror to check that data complete the self-adaptative adjustment of parameter by updating cumulative probability.

(5), in the step 5, rearview mirror checks that behavior asset pricing judges specifically to adopt with the following method：

Definition respectively stares coefficient when checking left and right rearview mirror：

μ_L=| φ-φ_L|/φ_L (9)

μ_R=| φ-φ_R|/φ_R (10)

Work as μ_L≤μ_L0When confirm that driver eye movement fixation point is located at vehicle left-hand mirror, work as μ_R≤μ_R0When confirm driver's eye Dynamic fixation point is located at vehicle right rear view mirror, wherein μ_L0And μ_R0Threshold value is stared respectively.Needs are checked in view of effective rearview mirror Persistently staring for certain time length, counts μ respectively_L≤μ_L0And μ_R≤μ_R0Continuous φ values number n_LAnd n_R, and if only if stares left back Duration T n of visor_L≥T_L0When judge driver observe left-hand mirror effectiveness, and if only if stares right rear view mirror Duration T n_R≥T_R0When judge driver observe right rear view mirror effectiveness.Wherein T_L0And T_R0Respectively persistent period Threshold value.

Preliminary experiment test shows that the recognition accuracy of above detection method can reach 85.6%.

In the present invention, each key parameter γ, Ω, ξ, r, ε, α, β, l, μ in image procossing and behavior decision algorithm_L0、μ_R0、 T_L0And T_R0Driver can be increased in system button is artificially adjusted according to practical situation debugging demarcation, also can be by vehicle manufacture Business or relevant authority department specify.

Additionally, the present invention can work independently in other electrical systems of vehicle, original connections, system peace are not destroyed Dress, maintenance and replacing are all relatively easy.

The above, is only presently preferred embodiments of the present invention, and not the technical scope of the present invention is imposed any restrictions, As long as therefore the change done of claim under this invention and description or modification, should all belong to the scope that patent of the present invention covers Within.

Claims (10)

1. a kind of driver's rearview mirror checks behavioral value method, it is characterised in that：With vehicle carrying video system T realities at timed intervals Shi Lianxu gathers driver's cabin image frame by frame, with the visible skin areas of driver face and cervical region collectively as image analysis processing Object, setting image processing algorithm come adaptive learning and tracking driver's face neck visible skin areas, extract driver's face Neck visible dermis outline, the cervical region minimum point with face neck visible dermis outline cross the vertical line of the basic point by face as basic point Neck visible dermis outline is divided into left and right two parts, defines left and right two parts area ratio and is characterized parameter, obtains this feature and joins Several reference characteristic value, by the numerical relation of current signature parameter and reference characteristic value and with reference to staring threshold value and persistent period Threshold value is determining whether that there is rearview mirror checks behavior.

2. a kind of driver's rearview mirror according to claim 1 checks behavioral value method, it is characterised in that：Open in vehicle The initial face neck visible skin areas of driver are recognized using frames differencing searching algorithm adaptive learning during dynamic.

3. a kind of driver's rearview mirror according to claim 1 checks behavioral value method, it is characterised in that：In vehicle row Driver's face neck visible skin areas during sailing based on prior image frame adopt the current face neck visible dermis of breathing method recognition and tracking Region.

4. a kind of driver's rearview mirror according to claim 1 checks behavioral value method, it is characterised in that：Using described The cumulative probability local peaking regularity of distribution of the characteristic parameter in rearview mirror checks action process is estimating reference characteristic value.

5. a kind of driver's rearview mirror according to claim 1 checks behavioral value method, it is characterised in that specifically include Following steps：

Step one：When a vehicle is started, using the visible skin of face neck that frames differencing searching algorithm self-adapting estimation driver is initial Skin region simultaneously calculates face neck skin initial gray average in the region；

Step 2：During vehicle is travelled, driver's cabin video system is whenever a frame new images of sampling, the driving based on prior image frame Member's face neck visible skin areas locating rectangle quickly recognizes current face neck visible skin areas using breathing method, while completing current Face neck skin gray average is estimated；If now vehicle is not turned to, system is in face neck region dynamic tracking state, and repeats Step 2；Otherwise continue step 3；

Step 3：In the face neck identification positioning region internal calculation contour feature parameter of present image；

Step 4：Local peaking is calculated according to the cumulative probability of face neck visible dermis outline characteristic parameter and estimates that profile benchmark is special Value indicative；

Step 5：Calculated according to the numerical relation of current outline characteristic parameter and reference characteristic value and stare coefficient, according to staring threshold Value condition statistics driver observes the persistent period of rearview mirror state, further according to duration threshold condition criterion driver Whether implement effective rearview mirror and check behavior；

Step 6：In the presence of judging that the effective rearview mirror of driver checks behavior, all spies during being checked using this Levy parameter and update cumulative distribution probability.

6. a kind of driver's rearview mirror according to claim 2 or 5 checks behavioral value method, it is characterised in that driver Initial face neck visible skin areas recognition methodss are as follows：

The first step：Frame difference source images are searched for and do difference；Two image P are searched for i.e. in initial learn image₀₁And P₀₂, this two Driver head's attitude of image has significant difference and hand gestures difference is not obvious, and does difference to the two field pictures, obtains Obtain frame difference result；

Frame difference source images concrete search procedure can be：

First, any two field pictures P acquired image during vehicle launch concentrated_uAnd P_vDifference is done, note frame difference result is D：

D (i, j)=| P_u(i,j)-P_v(i,j)| (1)

In formula (1), i and j is the row and column of pixel respectively；

Then, D is horizontally divided into into D₁And D₂Upper and lower two parts, the ration of division is γ；Calculate D₁And D₂Pixel grey scale accumulation ratio ω：

$\mathrm{\ω}=\underset{i=1}{\overset{\mathrm{\γ}W}{\Σ}}\underset{j=1}{\overset{H}{\Σ}}d(i,j)/\underset{i=\mathrm{\γ}W}{\overset{W}{\Σ}}\underset{j=1}{\overset{H}{\Σ}}d(i,j)---\left(2\right)$

In formula (2), d (i, j) represents the pixel of the i-th row jth row of image D, and W and H is respectively the width and height of image D；

Finally, coefficient of determination Ω, the D if ω >=Ω are given₁The time gradient information for containing is noticeably greater than D₂, i.e. head pose It is changed significantly, by P_uAnd P_vAs frame difference source images P₀₁And P₀₂, search completes；Otherwise continue random frame Difference search；If being arbitrarily still unsatisfactory for ω >=Ω after two field pictures difference, two field pictures when taking ω maximums are frame difference source Image；Now frames differencing result is D to note₀, D₀=| P₀₁-P₀₂|；

Second step：Frame difference result binaryzation；Note binaryzation result images are B₀, i.e.,：

$b_0(i,j)=\left\{\begin{array}{cc}0& d_0(i,j)\&GreaterEqual;\mathrm{\&xi;}\\ 255& d_0(i,j)<\mathrm{\&xi;}\end{array}\right.---\left(3\right)$

Wherein b₀(i, j) represents image B₀The i-th row jth row pixel, ξ is binary-state threshold；

3rd step：Tiny speckle and gossamer are removed；Image B is removed₀Tiny speckle and gossamer, obtain image A₀；

4th step：Face neck visible skin areas are positioned；Calculate image A₀Boundary rectangle R₀₁, which will be surrounded in frame difference source images Driver face, by rectangle R₀₁Certain multiple is extended downwardly highly for rectangle R₀₂, R₀₂Driver face and neck will be surrounded simultaneously will Portion, defines R₀₂For driver's face neck area-of-interest, that is, obtain described face neck visible skin areas；

5th step：Face neck skin gray average is calculated；Calculate area-of-interest R₀₂ε span grey level histograms, i.e., count respectively Gray value is located at interval pixel ratio, corresponds to the hair gray scale peak value or face of driver in rectangular histogram with two peak values respectively Neck skin gray scale peak value, selectes skin gray scale peak value g according to practical situation₀, and calculate [g₀-ε,g₀+ ε] between pixel grey scale it is equal Value can obtain face neck skin gray average

6th step：Initial face neck visible skin areas identification positioning；According toIn R₀₂Bilateral threshold filtering binaryzation are done in inside, The face neck pachylosiies image H ' of driver can be obtained₀：

$h_0^{\&prime;}(i,j)=\left\{\begin{array}{cc}0& s_{01}\&le;p_0(i,j)\&le;s_{02}\\ 255& p_0(i,j)<s_{01},p_0(i,j)>s_{02}\end{array}\right.---\left(4\right)$

Wherein：h′₀(i, j) represents image H₀' the i-th row jth row pixel,By H₀' repeat Three step process, filtering interfering obtain face neck skin image H₀, its boundary rectangle R₀As frame difference source images P₀₁In face neck can See skin area, i.e., initial face neck visible skin areas.

7. a kind of driver's rearview mirror according to claim 3 or 5 checks behavioral value method, it is characterised in that driver The domain breathing recognition methodss of face neck visible skin areas are as follows：

A two field picture is gathered every T time, note kth time sampled images are P_k, driver's face neck region be R_k, face neck skin gray scale Average isK >=1, P₀=P₀₁For initial condition, initial parameter R₀WithTried to achieve according to frame difference algorithm in vehicle launch；

The first step：Face neck area-of-interest is positioned；With R_k-1Center (x_k-1,y_k-1) for basic point, to R_k-1Expanse is carried out, is obtained Obtain present image P_kIn driver face neck area-of-interest R '_k, makeThe expansion factor of note width and height Respectively α and β, then：

w_k=w_k-1+Round(α·T)+2 (5)

h_k=h_k-1+Round(β·T)+2 (6)

Wherein w_k-1And h_k-1Respectively region R_k-1Width and height, w_kAnd h_kRespectively region R_k' width and height, Round (.) is bracket function；

Second step：Face neck skin gray average is calculated；Due to the face neck skin gray average in image in little sampling time interval T Change very little, therefore R '_kIt is located in the ε span grey level histograms in region Interval peak value g_kMust be The face neck skin gray scale peak value of present image, face neck skin gray averageAs [g_k-ε,g_k+ ε] interval pixel grey scale is equal Value；

3rd step：Face neck visible skin areas are recognized；In face neck area-of-interest R '_kInside, does bilateral threshold filtering two-value Change, extract the coarse image H ' of driver's face neck visible dermis_k：

$h_k^{\&prime;}(i,j)=\left\{\begin{array}{cc}0& s_{k1}\&le;p_k(i,j)\&le;s_{k2}\\ 255& p_k(i,j)<s_{k1},p_k(i,j)>s_{k2}\end{array}\right.---\left(7\right)$

Wherein h '_k(i, j) represents image H '_kThe i-th row jth row pixel,To H '_kCarry out Filter fleck and gossamer interference, the final binary image H for obtaining face neck visible dermis_k, its boundary rectangle R_kAs image P_kIn face neck visible skin areas.

8. a kind of driver's rearview mirror according to claim 1,4 or 5 checks behavioral value method, it is characterised in that face neck Visible dermis outline calculation of characteristic parameters and reference characteristic value evaluation method are as follows：

The closed curve of face neck visible dermis outline is obtained in the face neck visible skin areas of gained, first by cervical region low spot As reference mark O of contour line_k；

Secondly, cross O_kVertical line face neck outline is divided into into left and right two parts, its area S_kLAnd S_kRIt is relevant with head pose, area Than φ it is：

φ=S_kR/S_kL (8)

φ_L、φ₀And φ_RLeft and right area ratio when observation left-hand mirror, dead ahead and right rear view mirror is corresponded to respectively, defines three On the basis of eigenvalue；

The parameter phi and its accumulation effectively checked in action process by T sampled images interval time, the continuous l rearview mirror of calculating is general Rate, obtains φ using the local peaking that cumulative probability is distributed_L、φ₀And φ_REstimated value.

9. behavioral value method is checked according to a kind of arbitrary described driver's rearview mirror of claim 1-5, it is characterised in that after Visor checks that behavior asset pricing decision method is：

Definition respectively stares coefficient when checking left and right rearview mirror：

μ_L=| φ-φ_L|/φ_L (9)

μ_R=| φ-φ_R|/φ_R (10)

Work as μ_L≤μ_L0When confirm that driver eye movement fixation point is located at vehicle left-hand mirror, work as μ_R≤μ_R0When confirm driver eye movement coagulate Viewpoint is located at vehicle right rear view mirror, wherein μ_L0And μ_R0Threshold value is stared respectively；Statistics meets μ respectively_L≤μ_L0And μ_R≤μ_R0Condition Continuous sampling image number n_LAnd n_R, and if only if stares duration T n of left-hand mirror_L≥T_L0When judge driver The effectiveness of observation left-hand mirror, and if only if stares duration T n of right rear view mirror_R≥T_R0When judge driver observation The effectiveness of right rear view mirror, wherein T_L0And T_R0Respectively duration threshold.

10. a kind of driver's rearview mirror checks behavioral value system, it is characterised in that：Including system switching, power module, image Acquisition module and controller；The system switching is used to be turned on and off system；The power module is used to be that system and each module are supplied Electricity；The image capture module is for being acquired to driving indoor driver's image；The controller is to image capture module The image for collecting is processed, and judges whether driver effectively checks rearview mirror, and which includes frame difference evaluator, breathing identification Device, face neck contour feature parameter calculator, profile reference characteristic value estimator and rearview mirror check behavior asset pricing determinant；The frame Difference evaluator completes gray average of driver's face neck skin under the conditions of current light source and learns and initial in vehicle launch Face neck visible skin areas adaptable search is recognized；The breathing evaluator completes driver's face neck skin ash during vehicle traveling The study of degree average and the quick Tracking Recognition of face neck visible skin areas；The face neck contour feature parameter calculator is used to calculate current The face neck outline characteristic parameter of driver in the image of collection；The profile reference characteristic value estimator utilizes the face neck foreign steamer The cumulative probability of wide characteristic parameter is estimating profile reference characteristic value；The rearview mirror checks behavior asset pricing determinant according to current figure The face neck outline characteristic parameter of picture and profile reference characteristic value judge whether driver implements effective rearview mirror and check row For；The controller may further include, contour feature parameter cumulative probability renovator, and the contour feature parameter cumulative probability is more New device is used for the accumulation probability of occurrence of real-time update face neck contour feature parameter；Communication module is may further include, the communication The next Vehicular turn information of other electronic control units transmission of module reception vehicle, while will judge that conclusion is sent to other side and enters one Step is used.