CN104537649A - Vehicle steering judgment method and system based on image ambiguity comparison - Google Patents

Abstract

The invention provides a vehicle steering judgment method based on image ambiguity comparison. The method includes the steps of collecting driving images of a vehicle at all moments in real time through a camera shooting device, receiving the driving images and dividing the driving images into a plurality of sub-images, extracting the sub-images sensitive to gray level and color gradation changes to form ROI areas, symmetrically dividing each ROI area into a left area and a right area, calculating the image ambiguity degree of each area, counting the difference values between the image ambiguity degrees of the left areas and the right areas of the ROI areas composed of 300 or more driving images corresponding to different speeds, calculating mean values and variances of all the difference values at the corresponding speeds, and judging whether the vehicle is in the turning state or not and judging the steering direction of the vehicle according to the difference value between the image ambiguity degrees of the left area and the right area of the corresponding ROI area at the current speed of the vehicle. The method is small in calculation amount, the turning state of the vehicle can be rapidly judged in real time without being limited by day or night conditions, and the method is particularly suitable for back-mounted portable navigation products.

Description

A kind of Vehicular turn determination methods of comparing based on image blur and system

Technical field

The present invention relates to image processing field, more particularly, relate to a kind of Vehicular turn determination methods of comparing based on image blur and system.

Background technology

Along with the development of society, automobile is more and more welcomed by the general public, and increasing people has automobile.But due to the frequent generation of motor-vehicle accident, the driving safety of automobile but becomes the key issue that society be can not ignore.Motor-vehicle accident often occurs in turning, when line item instrument monitor turn inside diameter and excessive velocities time need produce report to the police, to carry out safety instruction to driver, particularly concerning portable rear dress navigation instrument, due to vehicle steering signal, wheel shaft turn signal etc. cannot be obtained easily as the original-pack equipment of vehicle, often increase and judge duration and therefore have little time escape from danger, so how to judge whether automobile is in turn condition in real time is fast a very important problem.

Whether traditional technique in measuring automobile is in turn condition primarily of following several: 1, rely on and detect steering indicating light electrical level judging, but this method cannot tackle the situation that human pilot does not play steering indicating light turning; 2, detection bearing circle or wheel steering device signal is relied on turning to of tire to be detected, but this method needs to carry out transformation and external to former car circuit, often to carry out brokenly line to former car circuit to install and could obtain signal, be unfavorable for the use of vehicle maintenance and rear dress portable navigation instrument; 3, utilize gps signal and gyroscope to carry out turning to judge, but this method is not only subject to satellite-signal impact but also needs the gyroscope that installs additional costly, can increase use cost.

Because common Tachographs or Tachographs navigation all-in-one machine are all with videograph camera, thus utilize camera collection to image information carry out turn inside diameter condition adjudgement fast, be a desirable solution route.The correlation technique of current existence, such as, on March 20th, 2013 is announced, notification number is the Chinese invention " judge the method for turning of mobile object and use the guider of the method " of 101782394, and described method comprises: the current position obtaining a mobile object according to gps receiver one of on a guider; Judge that whether this mobile object is apart from crossroad one predeterminable range; When mobile object is apart from this predeterminable range of this crossroad, image capturing device one of is utilized on this guider to take and obtain an initial crossing image; Utilize this image capturing device and take every a given time and obtain multiple continuous print crossings image; Calculate one of the moving direction of this object and this mobile object angle according to this initial crossing image and the plurality of continuous crossing image, and judge whether this mobile object completes a cornering operation according to the change of this angle.The program must carry out calculating to judge whether automobile turns in conjunction with GPS and map intersection information to the image collected, and the restriction being subject to satellite-signal impact and map datum real-time update makes judged result inaccurate.

And for example, disclosed in 14 days November in 2012, publication number is the Chinese invention " a kind of method judging vehicle running state " of 102774380, described method comprises, in the composograph of vehicle-mounted viewing system, the characteristics of image that Automatic-searching can be followed the trail of, and carry out characteristic matching in the next frame, obtain the mobile vector of each feature; Vehicle movement and sense of rotation is obtained by the feature mobile vector COMPREHENSIVE CALCULATING after each coupling; Vehicle running state is judged by the displacement and the anglec of rotation that calculate front and back frame vehicle.The program carries out turn condition judgement by the method for mating the composograph of vehicle-mounted viewing system, large owing to looking around image data amount, matching algorithm calculated amount is high, can consume a large amount of computational resources and speed, thus causing the overlong time judging turn condition, practicality is not high.

ROI (region of interest, area-of-interest), in machine vision, image procossing, sketches the contours of need region to be processed from processed image in modes such as square frame, circle, ellipse, irregular polygons, is called area-of-interest.In image processing field, area-of-interest is the image-region selected from image, and this region is the emphasis that your graphical analysis is paid close attention to, and draws a circle to approve this region to be further processed, and can reduce the processing time like this, increases precision.

Summary of the invention

One of the technical problem to be solved in the present invention, be to provide a kind of Vehicular turn determination methods compared based on image blur, according to Ackermann steer angle, outside wheel speed is faster than inboard wheel speed, and the image that therefore picture pick-up device gathers is greater than turning medial area pixel blur level in turning exterior lateral area pixel blur level.By carrying out quantification to the traveling left ROI region of image and the image blur of right ROI region and compare, thus judge whether vehicle is in turn condition and turn direction.Calculated amount is little, can carry out turn condition judgement real-time to vehicle, and not by the restriction of day and night condition, is specially adapted to rear dress portable navigation product.

One of the present invention is achieved in that a kind of Vehicular turn determination methods compared based on image blur, and described method comprises: image acquisition process, image segmentation process, blur level computation process, blur level statistic processes and turn to deterministic process:

Described image acquisition process: the traveling image in picture pick-up device Real-time Collection vehicle each moment;

Described image segmentation process: receive described traveling image and be a plurality of subimage by described traveling Iamge Segmentation, deletes wherein gray scale and color range and changes mild subimage, thus extracts the subimage composition ROI region of gray scale and color range sensitive;

Described blur level computation process: the ROI region symmetry of the described traveling image obtained is divided into left ROI region and right ROI region, adopts horizontal gray scale average gradient algorithm to calculate the image blur of left ROI region and the image blur of right ROI region respectively;

Described blur level statistic processes: in conjunction with the velocity information of vehicle, count the difference travelling image blur between the left ROI region of image and right ROI region described at least 300 width corresponding with friction speed respectively, then obtained by each mathematic interpolation image blur between corresponding speed bottom left ROI region and right ROI region difference mean and variance and be kept at local real-time update;

Describedly turn to deterministic process: by the difference of image blur between vehicle current time speed bottom left ROI region and right ROI region, and the mean and variance of the difference of image blur between this speed bottom left ROI region and right ROI region preserved, judge that vehicle is current and whether be in turn condition and turn direction, and export corresponding signal for turn.

Preferably, described picture pick-up device is placed in the middle place of shield glass.

Preferably, described image segmentation process is specially further:

Step 10, be the subimage of continuous nonoverlapping N*N pixel by the described traveling Iamge Segmentation received; Described N is positive integer and is determined by the resolution of picture pick-up device: resolution is higher, and N is less;

Step 11, calculate the entropy of each subimage respectively, and respectively the entropy of each subimage and first threshold are compared: if entropy is greater than first threshold, then represent gray scale and the color range sensitive of this subimage; If entropy is less than first threshold, then represent that the gray scale of this subimage and color range change mild; Described first threshold is the empirical value obtained by experiment;

Step 12, deletion gray scale and color range change mild subimage, extract the subimage composition ROI region of each gray scale and color range sensitive.

Preferably, the formula of described horizontal gray scale average gradient algorithm is as follows:

${\mathrm{GMC}}_{L/R}=\frac{\underset{i=1}{\overset{I-1}{\mathrm{\Σ}}}\underset{j=1}{\overset{J-1}{\mathrm{\Σ}}}|r(i+1,j)-r(i,j)|+|g(i+1,j)-g(i,j)|+|b(i+1,j)-b(i,j)|}{3*\mathrm{CountROI}}---\left(1\right)$

In formula (1), (i+1, j) ∈ ROI & & (i, j) ∈ ROI, j is the height of described traveling image, i is the half of described traveling picture traverse, r (i, j) be the pixel value of described traveling image red component, g (i, j) is the pixel value of described traveling image green component, b (i, j) be the pixel value of described traveling image blue component, CountROI is the number of left ROI region or right ROI region pixel, GMC _lrepresent the image blur of left ROI region, GMC _rrepresent the image blur of right ROI region;

Above-mentioned each value is substituted into formula (1) respectively, and just can obtain corresponding left ROI region or the GMC value of right ROI region, GMC value larger expression image frame is more clear, and GMC value less expression image frame is fuzzyyer.

Preferably, deterministic process is turned to be specially further described in:

The image blur GMC of the left ROI region of described traveling image under the current time speed that step 20, reception calculate _lwith the image blur GMC of right ROI region _rand the difference M of both correspondences;

Step 21, inquiry is local whether has the mean and variance of the difference of image blur between vehicle present speed bottom left ROI region and right ROI region for calling: if so, then jump procedure 22; Then return step 20 if not;

Step 22, calculating Second Threshold Th=e+3 σ, wherein e represents the average of the difference of image blur between present speed bottom left ROI region and right ROI region, and σ represents the variance of the difference of image blur between present speed bottom left ROI region and right ROI region;

Step 23, judge whether M is more than or equal to Th: if so, then represent that vehicle is current and be in turn condition, jump procedure 24; If not, then represent that vehicle is current and be not in turn condition, do not carry out signal for turn output;

Step 24: judge GMC _lwhether be greater than GMC _r: if so, then represent that the image frame of right ROI region is fuzzyyer, namely right side wheels is in outside bend, and vehicle is current is turn left, exports left turn signal; If not, then represent that the image frame of left ROI region is fuzzyyer, namely left side wheel is in outside bend, and vehicle is current is bend to right, and exports right turn signal.

The technical problem to be solved in the present invention two, be to provide a kind of Vehicular turn compared based on image blur to judge system, according to Ackermann steer angle, outside wheel speed is faster than inboard wheel speed, and the image that therefore picture pick-up device gathers is greater than turning medial area pixel blur level in turning exterior lateral area pixel blur level.By carrying out quantification to the traveling left ROI region of image and the image blur of right ROI region and compare, thus judge whether vehicle is in turn condition and turn direction.Calculated amount is little, can carry out turn condition judgement real-time to vehicle, and not by the restriction of day and night condition, is specially adapted to rear dress portable navigation product.

The present invention's two is achieved in that a kind of Vehicular turn compared based on image blur judges system, and described system comprises: image capture module, image segmentation module, blur level computing module, blur level statistical module and turn to judge module:

Described image capture module: for the traveling image in picture pick-up device Real-time Collection vehicle each moment;

Described image segmentation module: for receiving described traveling image and be a plurality of subimage by described traveling Iamge Segmentation, deletes wherein gray scale and color range and changes mild subimage, thus extract the subimage composition ROI region of gray scale and color range sensitive;

Described blur level computing module: for the ROI region symmetry of the described traveling image obtained is divided into left ROI region and right ROI region, adopts horizontal gray scale average gradient algorithm to calculate the image blur of left ROI region and the image blur of right ROI region respectively;

Described blur level statistical module: for the velocity information in conjunction with vehicle, count the difference travelling image blur between the left ROI region of image and right ROI region described at least 300 width corresponding with friction speed respectively, then obtained by each mathematic interpolation image blur between corresponding speed bottom left ROI region and right ROI region difference mean and variance and be kept at local real-time update;

Describedly turn to judge module: for the difference by image blur between vehicle current time speed bottom left ROI region and right ROI region, and the mean and variance of the difference of image blur between this speed bottom left ROI region and right ROI region preserved, judge that vehicle is current and whether be in turn condition and turn direction, and export corresponding signal for turn.

Preferably, described picture pick-up device is placed in the middle place of shield glass.

Preferably, described image segmentation module comprises extraction unit, and for splitting described traveling image and extracting ROI region, detailed process is as follows:

Step 10, be the subimage of continuous nonoverlapping N*N pixel by the described traveling Iamge Segmentation received; Described N is positive integer and is determined by the resolution of picture pick-up device: resolution is higher, and N is less;

Step 11, calculate the entropy of each subimage respectively, and respectively the entropy of each subimage and first threshold are compared: if entropy is greater than first threshold, then represent gray scale and the color range sensitive of this subimage; If entropy is less than first threshold, then represent that the gray scale of this subimage and color range change mild; Described first threshold is the empirical value obtained by experiment;

Step 12, deletion gray scale and color range change mild subimage, extract the subimage composition ROI region of each gray scale and color range sensitive.

Preferably, the formula of described horizontal gray scale average gradient algorithm is as follows:

${\mathrm{GMC}}_{L/R}=\frac{\underset{i=1}{\overset{I-1}{\mathrm{\Σ}}}\underset{j=1}{\overset{J-1}{\mathrm{\Σ}}}|r(i+1,j)-r(i,j)|+|g(i+1,j)-g(i,j)|+|b(i+1,j)-b(i,j)|}{3*\mathrm{CountROI}}---\left(1\right)$

In formula (1), (i+1, j) ∈ ROI & & (i, j) ∈ ROI, j is the height of described traveling image, i is the half of described traveling picture traverse, r (i, j) be the pixel value of described traveling image red component, g (i, j) is the pixel value of described traveling image green component, b (i, j) be the pixel value of described traveling image blue component, CountROI is the number of left ROI region or right ROI region pixel, GMC _lrepresent the image blur of left ROI region, GMC _rrepresent the image blur of right ROI region;

Above-mentioned each value is substituted into formula (1) respectively, and just can obtain corresponding left ROI region or the GMC value of right ROI region, GMC value larger expression image frame is more clear, and GMC value less expression image frame is fuzzyyer.

Preferably, described in turn to judge module to comprise analytic unit, for calculating and analyzing turn condition and the turn direction of vehicle, detailed process is:

The image blur GMC of the left ROI region of described traveling image under the current time speed that step 20, reception calculate _lwith the image blur GMC of right ROI region _rand the difference M of both correspondences;

Step 21, inquiry is local whether has the mean and variance of the difference of image blur between vehicle present speed bottom left ROI region and right ROI region for calling: if so, then jump procedure 22; Then return step 20 if not;

Step 22, calculating Second Threshold Th=e+3 σ, wherein e represents the average of the difference of image blur between present speed bottom left ROI region and right ROI region, and σ represents the variance of the difference of image blur between present speed bottom left ROI region and right ROI region;

Step 23, judge whether M is more than or equal to Th: if so, then represent that vehicle is current and be in turn condition, jump procedure 24; If not, then represent that vehicle is current and be not in turn condition, do not carry out signal for turn output;

Step 24: judge GMC _lwhether be greater than GMC _r: if so, then represent that the image frame of right ROI region is fuzzyyer, namely right side wheels is in outside bend, and vehicle is current is turn left, exports left turn signal; If not, then represent that the image frame of left ROI region is fuzzyyer, namely left side wheel is in outside bend, and vehicle is current is bend to right, and exports right turn signal.

After adopting such scheme, tool of the present invention has the following advantages:

1, by image blur computing method, the image blur of the left ROI region of image and right ROI region is carried out quantification and compared, thus judge whether vehicle is in turn condition and turn direction, and calculated amount is little, turn condition easily can be carried out to vehicle real-time and judge;

2, picture pick-up device is arranged on the middle of shield glass, the scope making collected traveling image the right and left cause blur level to change by velocity contrast is like this identical with degree, can ensure the accuracy judged;

3, by the segmentation to traveling image, and the region extracting gray scale and color range sensitive in subimage is analyzed as ROI region, and so targetedly to region analysis, can reduce image processing time, purpose is stronger;

4, horizontal gray scale average gradient algorithm is adopted to calculate the image blur of left ROI region and right ROI region, landscape blur degree can be caused to change because vehicle travels general, so need compared to traditional gray scale average gradient algorithm the shade of gray calculating four direction, decrease the calculated amount of image blur, improve judgement speed;

Under the friction speed of 5, preserving this locality, between corresponding left ROI region and right ROI region, the mean and variance of the difference of image blur upgrades, and ensure that real-time and the accuracy of data;

6, due to vehicle image blur rate of change meeting difference to some extent at various speeds itself, turning the blur level difference scope at various speeds caused also can be different, by adding up the mean and variance of the image blur difference of friction speed bottom left ROI region and right ROI region, can be used for and turn to judge module to make correct Vehicular turn judgement with reference to decision-making.

Accompanying drawing explanation

The present invention is further illustrated in conjunction with the embodiments with reference to the accompanying drawings.

Fig. 1 is the inventive method flowchart.

Fig. 2 be the inventive method one embodiment turn to deterministic process process flow diagram.

Fig. 3 is the connection diagram of present system.

Embodiment

Refer to Fig. 1, the present invention, a kind of Vehicular turn determination methods compared based on image blur, described method comprises: image acquisition process, image segmentation process, blur level computation process, blur level statistic processes and turn to deterministic process:

Described image acquisition process: the traveling image in picture pick-up device Real-time Collection vehicle each moment;

Described image segmentation process: receive described traveling image and be a plurality of subimage by described traveling Iamge Segmentation, delete wherein gray scale and color range and change mild subimage (sky that such as color is the same, road surface, night, at a distance black region etc. all belonged to gray scale and color range changes mild subimage, they are insensitive to the smear out effect of image), thus extract the subimage composition ROI region of gray scale and color range sensitive; Like this can pointedly to region analysis, reduce image processing time, purpose is stronger;

Described blur level computation process: the ROI region symmetry of the described traveling image obtained is divided into left ROI region and right ROI region, adopts horizontal gray scale average gradient algorithm to calculate the image blur of left ROI region and the image blur of right ROI region respectively;

Described blur level statistic processes: in conjunction with the velocity information of vehicle, count the difference travelling image blur between the left ROI region of image and right ROI region described at least 300 width corresponding with friction speed respectively, then obtained by each mathematic interpolation image blur between corresponding speed bottom left ROI region and right ROI region difference mean and variance and be kept at local real-time update; Ensure that real-time and the accuracy of data.Such as, automobile is current with the speeds of 60Km/h, during automobile is with this speeds, then counts the difference of image blur between the left ROI region of at least 300 width traveling images that picture pick-up device collects and right ROI region, and goes out mean and variance by these mathematic interpolation.

Describedly turn to deterministic process: by the difference of image blur between vehicle current time speed bottom left ROI region and right ROI region, and the mean and variance of the difference of image blur between this speed bottom left ROI region and right ROI region preserved, judge that vehicle is current and whether be in turn condition and turn direction, and export corresponding signal for turn.

Pass through said process, subimage after extracting segmentation adopts horizontal gray scale average gradient algorithm to the image blur of the left ROI region and right ROI region that calculate described traveling image, and difference comparsion is carried out to the image blur of the left ROI region of image and right ROI region, can judge whether vehicle is in turn condition and turn direction;

Described picture pick-up device is placed in the middle place of shield glass; The scope making collected traveling image the right and left cause blur level to change by velocity contrast is like this identical with degree, can ensure the accuracy judged.

Described image segmentation process is specially further:

Step 10, be the subimage of continuous nonoverlapping N*N pixel by the described traveling Iamge Segmentation received; Described N is positive integer and is determined by the resolution of picture pick-up device: resolution is higher, and N is less; Such as, for the resolution of 1920*1080, N desirable 20;

Step 11, calculate the entropy of each subimage respectively, and respectively the entropy of each subimage and first threshold are compared: if entropy is greater than first threshold, then represent gray scale and the color range sensitive of this subimage; If entropy is less than first threshold, then represent that the gray scale of this subimage and color range change mild; Described first threshold is the empirical value obtained by experiment;

Step 12, deletion gray scale and color range change mild subimage, extract the subimage composition ROI region of each gray scale and color range sensitive.

The formula of described horizontal gray scale average gradient algorithm is as follows:

${\mathrm{GMC}}_{L/R}=\frac{\underset{i=1}{\overset{I-1}{\mathrm{\Σ}}}\underset{j=1}{\overset{J-1}{\mathrm{\Σ}}}|r(i+1,j)-r(i,j)|+|g(i+1,j)-g(i,j)|+|b(i+1,j)-b(i,j)|}{3*\mathrm{CountROI}}---\left(1\right)$

In formula (1), (i+1, j) ∈ ROI & & (i, j) ∈ ROI, j is the height of described traveling image, i is the half of described traveling picture traverse, r (i, j) be the pixel value of described traveling image red component, g (i, j) is the pixel value of described traveling image green component, b (i, j) be the pixel value of described traveling image blue component, CountROI is the number of left ROI region or right ROI region pixel, GMC _lrepresent the image blur of left ROI region, GMC _rrepresent the image blur of right ROI region;

Above-mentioned each value is substituted into formula (1) respectively, and just can obtain corresponding left ROI region or the GMC value of right ROI region, GMC value larger expression image frame is more clear, and GMC value less expression image frame is fuzzyyer; Adopt horizontal gray scale average gradient algorithm, need compared to traditional gray scale average gradient algorithm the shade of gray calculating four direction, decrease the calculated amount of image blur, improve judgement speed.

As shown in Figure 3, deterministic process is turned to be specially further described in:

The image blur GMC of the left ROI region of described traveling image under the current time speed that step 20, reception calculate _lwith the image blur GMC of right ROI region _rand the difference M of both correspondences;

Step 21, inquiry is local whether has the mean and variance of the difference of image blur between vehicle present speed bottom left ROI region and right ROI region for calling: if so, then jump procedure 22; Then return step 20 if not;

Step 22, calculating Second Threshold Th=e+3 σ, wherein e represents the average of the difference of image blur between present speed bottom left ROI region and right ROI region, and σ represents the variance of the difference of image blur between present speed bottom left ROI region and right ROI region;

Step 23, judge whether M is more than or equal to Th: if so, then represent that vehicle is current and be in turn condition, jump procedure 24; If not, then represent that vehicle is current and be not in turn condition, do not carry out signal for turn output;

Step 24: judge GMC _lwhether be greater than GMC _r: if so, then represent that the image frame of right ROI region is fuzzyyer, namely right side wheels is in outside bend, and vehicle is current is turn left, exports left turn signal; If not, then represent that the image frame of left ROI region is fuzzyyer, namely left side wheel is in outside bend, and vehicle is current is bend to right, and exports right turn signal.

Based on above method, as shown in Figure 3, the present invention, a kind of Vehicular turn compared based on image blur judges system, and described system comprises: image capture module, image segmentation module, blur level computing module, blur level statistical module and turn to judge module:

Described image capture module: for the traveling image in picture pick-up device Real-time Collection vehicle each moment;

Described image segmentation module: for receiving described traveling image and be a plurality of subimage by described traveling Iamge Segmentation, deletes wherein gray scale and color range and changes mild subimage, thus extract the subimage composition ROI region of gray scale and color range sensitive; Like this can pointedly to region analysis, reduce image processing time, purpose is stronger;

Described blur level computing module: for the ROI region symmetry of the described traveling image obtained is divided into left ROI region and right ROI region, adopts horizontal gray scale average gradient algorithm to calculate the image blur of left ROI region and the image blur of right ROI region respectively;

Described blur level statistical module: for the velocity information in conjunction with vehicle, count the difference travelling image blur between the left ROI region of image and right ROI region described at least 300 width corresponding with friction speed respectively, then obtained by each mathematic interpolation image blur between corresponding speed bottom left ROI region and right ROI region difference mean and variance and be kept at local real-time update; Ensure that real-time and the accuracy of data.

Describedly turn to judge module: for the difference by image blur between vehicle current time speed bottom left ROI region and right ROI region, and the mean and variance of the difference of image blur between this speed bottom left ROI region and right ROI region preserved, judge that vehicle is current and whether be in turn condition and turn direction, and export corresponding signal for turn.

Described picture pick-up device is placed in the middle place of shield glass; The scope making collected traveling image the right and left cause blur level to change by velocity contrast is like this identical with degree, can ensure the accuracy judged.

Described image segmentation module comprises extraction unit, and for splitting described traveling image and extracting ROI region, detailed process is as follows:

Step 10, be the subimage of continuous nonoverlapping N*N pixel by the described traveling Iamge Segmentation received; Described N is positive integer and is determined by the resolution of picture pick-up device: resolution is higher, and N is less;

Step 11, calculate the entropy of each subimage respectively, and respectively the entropy of each subimage and first threshold are compared: if entropy is greater than first threshold, then represent gray scale and the color range sensitive of this subimage; If entropy is less than first threshold, then represent that the gray scale of this subimage and color range change mild; Described first threshold is the empirical value obtained by experiment;

Step 12, deletion gray scale and color range change mild subimage, extract the subimage composition ROI region of each gray scale and color range sensitive.

The formula of described horizontal gray scale average gradient algorithm is as follows:

${\mathrm{GMC}}_{L/R}=\frac{\underset{i=1}{\overset{I-1}{\mathrm{\Σ}}}\underset{j=1}{\overset{J-1}{\mathrm{\Σ}}}|r(i+1,j)-r(i,j)|+|g(i+1,j)-g(i,j)|+|b(i+1,j)-b(i,j)|}{3*\mathrm{CountROI}}---\left(1\right)$

In formula (1), (i+1, j) ∈ ROI & & (i, j) ∈ ROI, j is the height of described traveling image, i is the half of described traveling picture traverse, r (i, j) be the pixel value of described traveling image red component, g (i, j) is the pixel value of described traveling image green component, b (i, j) be the pixel value of described traveling image blue component, CountROI is the number of left ROI region or right ROI region pixel, GMC _lrepresent the image blur of left ROI region, GMC _rrepresent the image blur of right ROI region;

Above-mentioned each value is substituted into formula (1) respectively, and just can obtain corresponding left ROI region or the GMC value of right ROI region, GMC value larger expression image frame is more clear, and GMC value less expression image frame is fuzzyyer; Adopt horizontal gray scale average gradient algorithm, need compared to traditional gray scale average gradient algorithm the shade of gray calculating four direction, decrease the calculated amount of image blur, improve judgement speed.

The described judge module that turns to comprises analytic unit, and for calculating and analyzing turn condition and the turn direction of vehicle, detailed process is:

The image blur GMC of the left ROI region of described traveling image under the current time speed that step 20, reception calculate _lwith the image blur GMC of right ROI region _rand the difference M of both correspondences;

Step 21, inquiry is local whether has the mean and variance of the difference of image blur between vehicle present speed bottom left ROI region and right ROI region for calling: if so, then jump procedure 22; Then return step 20 if not;

Step 22, calculating Second Threshold Th=e+3 σ, wherein e represents the average of the difference of image blur between present speed bottom left ROI region and right ROI region, and σ represents the variance of the difference of image blur between present speed bottom left ROI region and right ROI region;

Step 23, judge whether M is more than or equal to Th: if so, then represent that vehicle is current and be in turn condition, jump procedure 24; If not, then represent that vehicle is current and be not in turn condition, do not carry out signal for turn output;

Step 24: judge GMC _lwhether be greater than GMC _r: if so, then represent that the image frame of right ROI region is fuzzyyer, namely right side wheels is in outside bend, and vehicle is current is turn left, exports left turn signal; If not, then represent that the image frame of left ROI region is fuzzyyer, namely left side wheel is in outside bend, and vehicle is current is bend to right, and exports right turn signal.

The present invention, according to Ackermann steer angle, outside wheel speed is faster than inboard wheel speed, and the traveling image that therefore picture pick-up device gathers is greater than turning medial area pixel blur level in turning exterior lateral area pixel blur level.By carrying out quantification to the image blur travelling the left ROI region that extracts of Image Segmentation Using and right ROI region and compare, thus judge whether vehicle is in turn condition and turn direction.Calculated amount is little, can carry out turn condition easily real-time and judge, and not by the restriction of day and night condition, be specially adapted to rear dress portable navigation product to vehicle.

Although the foregoing describe the specific embodiment of the present invention; but be familiar with those skilled in the art to be to be understood that; specific embodiment described by us is illustrative; instead of for the restriction to scope of the present invention; those of ordinary skill in the art, in the modification of the equivalence done according to spirit of the present invention and change, should be encompassed in scope that claim of the present invention protects.

Claims (10)

1. based on the Vehicular turn determination methods that image blur compares, it is characterized in that: described method comprises: image acquisition process, image segmentation process, blur level computation process, blur level statistic processes and turn to deterministic process:

Described image acquisition process: the traveling image in picture pick-up device Real-time Collection vehicle each moment;

Described image segmentation process: receive described traveling image and be a plurality of subimage by described traveling Iamge Segmentation, deletes wherein gray scale and color range and changes mild subimage, thus extracts the subimage composition ROI region of gray scale and color range sensitive;

Described blur level computation process: the ROI region symmetry of the described traveling image obtained is divided into left ROI region and right ROI region, adopts horizontal gray scale average gradient algorithm to calculate the image blur of left ROI region and the image blur of right ROI region respectively;

Described blur level statistic processes: in conjunction with the velocity information of vehicle, count the difference travelling image blur between the left ROI region of image and right ROI region described at least 300 width corresponding with friction speed respectively, then obtained by each mathematic interpolation image blur between corresponding speed bottom left ROI region and right ROI region difference mean and variance and be kept at local real-time update;

Describedly turn to deterministic process: by the difference of image blur between vehicle current time speed bottom left ROI region and right ROI region, and the mean and variance of the difference of image blur between this speed bottom left ROI region and right ROI region preserved, judge that vehicle is current and whether be in turn condition and turn direction, and export corresponding signal for turn.

2. a kind of Vehicular turn determination methods compared based on image blur according to claim 1, is characterized in that: described picture pick-up device is placed in the middle place of shield glass.

3. a kind of Vehicular turn determination methods compared based on image blur according to claim 1, is characterized in that: described image segmentation process is specially further:

Step 10, be the subimage of continuous nonoverlapping N*N pixel by the described traveling Iamge Segmentation received; Described N is positive integer and is determined by the resolution of picture pick-up device: resolution is higher, and N is less;

Step 11, calculate the entropy of each subimage respectively, and respectively the entropy of each subimage and first threshold are compared: if entropy is greater than first threshold, then represent gray scale and the color range sensitive of this subimage; If entropy is less than first threshold, then represent that the gray scale of this subimage and color range change mild; Described first threshold is the empirical value obtained by experiment;

Step 12, deletion gray scale and color range change mild subimage, extract the subimage composition ROI region of each gray scale and color range sensitive.

4. a kind of Vehicular turn determination methods compared based on image blur according to claim 1, is characterized in that: the formula of described horizontal gray scale average gradient algorithm is as follows:

${\mathrm{GMC}}_{L/R}=\frac{\underset{i=1}{\overset{I-1}{\mathrm{\Σ}}}\underset{j=1}{\overset{J-1}{\mathrm{\Σ}}}|r(i+1,j)-r(i,j)|+|g(i+1,j)-g(i,j)|+|b(i+1,j)-b(i-j)|}{3*\mathrm{CountROI}}---\left(1\right)$

In formula (1), (i+1, j) ∈ ROI & & (i, j) ∈ ROI, j is the height of described traveling image, i is the half of described traveling picture traverse, r (i, j) be the pixel value of described traveling image red component, g (i, j) is the pixel value of described traveling image green component, b (i, j) be the pixel value of described traveling image blue component, CountROI is the number of left ROI region or right ROI region pixel, GMC _lrepresent the image blur of left ROI region, GMC _rrepresent the image blur of right ROI region;

Above-mentioned each value is substituted into formula (1) respectively, and just can obtain corresponding left ROI region or the GMC value of right ROI region, GMC value larger expression image frame is more clear, and GMC value less expression image frame is fuzzyyer.

5. a kind of Vehicular turn determination methods compared based on image blur according to claim 1, is characterized in that: described in turn to deterministic process to be specially further:

The image blur GMC of the left ROI region of described traveling image under the current time speed that step 20, reception calculate _lwith the image blur GMC of right ROI region _rand the difference M of both correspondences;

Step 21, inquiry is local whether has the mean and variance of the difference of image blur between vehicle present speed bottom left ROI region and right ROI region for calling: if so, then jump procedure 22; Then return step 20 if not;

Step 22, calculating Second Threshold Th=e+3 σ, wherein e represents the average of the difference of image blur between present speed bottom left ROI region and right ROI region, and σ represents the variance of the difference of image blur between present speed bottom left ROI region and right ROI region;

Step 23, judge whether M is more than or equal to Th: if so, then represent that vehicle is current and be in turn condition, jump procedure 24; If not, then represent that vehicle is current and be not in turn condition, do not carry out signal for turn output;

Step 24: judge GMC _lwhether be greater than GMC _r: if so, then represent that the image frame of right ROI region is fuzzyyer, namely right side wheels is in outside bend, and vehicle is current is turn left, exports left turn signal; If not, then represent that the image frame of left ROI region is fuzzyyer, namely left side wheel is in outside bend, and vehicle is current is bend to right, and exports right turn signal.

6. the Vehicular turn compared based on image blur judges a system, it is characterized in that: described system comprises: image capture module, image segmentation module, blur level computing module, blur level statistical module and turn to judge module:

Described image capture module: for the traveling image in picture pick-up device Real-time Collection vehicle each moment;

Described image segmentation module: for receiving described traveling image and be a plurality of subimage by described traveling Iamge Segmentation, deletes wherein gray scale and color range and changes mild subimage, thus extract the subimage composition ROI region of gray scale and color range sensitive;

Described blur level computing module: for the ROI region symmetry of the described traveling image obtained is divided into left ROI region and right ROI region, adopts horizontal gray scale average gradient algorithm to calculate the image blur of left ROI region and the image blur of right ROI region respectively;

Described blur level statistical module: for the velocity information in conjunction with vehicle, count the difference travelling image blur between the left ROI region of image and right ROI region described at least 300 width corresponding with friction speed respectively, then obtained by each mathematic interpolation image blur between corresponding speed bottom left ROI region and right ROI region difference mean and variance and be kept at local real-time update;

Describedly turn to judge module: for the difference by image blur between vehicle current time speed bottom left ROI region and right ROI region, and the mean and variance of the difference of image blur between this speed bottom left ROI region and right ROI region preserved, judge that vehicle is current and whether be in turn condition and turn direction, and export corresponding signal for turn.

7. a kind of Vehicular turn compared based on image blur according to claim 6 judges system, it is characterized in that: described picture pick-up device is placed in the middle place of shield glass.

8. a kind of Vehicular turn compared based on image blur according to claim 6 judges system, it is characterized in that: described image segmentation module comprises extraction unit, and for splitting described traveling image and extracting ROI region, detailed process is as follows:

Step 10, be the subimage of continuous nonoverlapping N*N pixel by the described traveling Iamge Segmentation received; Described N is positive integer and is determined by the resolution of picture pick-up device: resolution is higher, and N is less;

Step 11, calculate the entropy of each subimage respectively, and respectively the entropy of each subimage and first threshold are compared: if entropy is greater than first threshold, then represent gray scale and the color range sensitive of this subimage; If entropy is less than first threshold, then represent that the gray scale of this subimage and color range change mild; Described first threshold is the empirical value obtained by experiment;

Step 12, deletion gray scale and color range change mild subimage, extract the subimage composition ROI region of each gray scale and color range sensitive.

9. a kind of Vehicular turn compared based on image blur according to claim 6 judges system, it is characterized in that: the formula of described horizontal gray scale average gradient algorithm is as follows:

${\mathrm{GMC}}_{L/R}=\frac{\underset{i=1}{\overset{I-1}{\mathrm{\Σ}}}\underset{j=1}{\overset{J-1}{\mathrm{\Σ}}}|r(i+1,j)-r(i,j)|+|g(i+1,j)-g(i,j)|+|b(i+1,j)-b(i-j)|}{3*\mathrm{CountROI}}---\left(1\right)$

In formula (1), (i+1, j) ∈ ROI & & (i, j) ∈ ROI, j is the height of described traveling image, i is the half of described traveling picture traverse, r (i, j) be the pixel value of described traveling image red component, g (i, j) is the pixel value of described traveling image green component, b (i, j) be the pixel value of described traveling image blue component, CountROI is the number of left ROI region or right ROI region pixel, GMC _lrepresent the image blur of left ROI region, GMC _rrepresent the image blur of right ROI region;

Above-mentioned each value is substituted into formula (1) respectively, and just can obtain corresponding left ROI region or the GMC value of right ROI region, GMC value larger expression image frame is more clear, and GMC value less expression image frame is fuzzyyer.

10. a kind of Vehicular turn compared based on image blur according to claim 6 judges system, it is characterized in that: described in turn to judge module to comprise analytic unit, for calculating and analyzing turn condition and the turn direction of vehicle, detailed process is:

The image blur GMC of the left ROI region of described traveling image under the current time speed that step 20, reception calculate _lwith the image blur GMC of right ROI region _rand the difference M of both correspondences;

Step 21, inquiry is local whether has the mean and variance of the difference of image blur between vehicle present speed bottom left ROI region and right ROI region for calling: if so, then jump procedure 22; Then return step 20 if not;

Step 22, calculating Second Threshold Th=e+3 σ, wherein e represents the average of the difference of image blur between present speed bottom left ROI region and right ROI region, and σ represents the variance of the difference of image blur between present speed bottom left ROI region and right ROI region;

Step 23, judge whether M is more than or equal to Th: if so, then represent that vehicle is current and be in turn condition, jump procedure 24; If not, then represent that vehicle is current and be not in turn condition, do not carry out signal for turn output;

Step 24: judge GMC _lwhether be greater than GMC _r: if so, then represent that the image frame of right ROI region is fuzzyyer, namely right side wheels is in outside bend, and vehicle is current is turn left, exports left turn signal; If not, then represent that the image frame of left ROI region is fuzzyyer, namely left side wheel is in outside bend, and vehicle is current is bend to right, and exports right turn signal.