CN102975659A - Automobile night driving pre-warning system and method based on far infrared image - Google Patents
Automobile night driving pre-warning system and method based on far infrared image Download PDFInfo
- Publication number
- CN102975659A CN102975659A CN2012104548976A CN201210454897A CN102975659A CN 102975659 A CN102975659 A CN 102975659A CN 2012104548976 A CN2012104548976 A CN 2012104548976A CN 201210454897 A CN201210454897 A CN 201210454897A CN 102975659 A CN102975659 A CN 102975659A
- Authority
- CN
- China
- Prior art keywords
- image
- human body
- gradient
- far infrared
- automobile
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Image Processing (AREA)
Abstract
The invention relates to an automobile night driving pre-warning system, in particular to composition and an achieving method of the automobile night driving pre-warning system based on monocular vision. The method can enable an automobile to remind a driver or directly take brake measures when the automobile probably collides. The system consists of hardware and software, wherein the hardware is composed of a far infrared image collecting device, a storage device, a processor, a display device, a warning device and a brake device. The software is composed of an image processing program and a control program. The achieving method includes firstly confirming by return on investment (ROI), direction gradient histogram, judgment classification, target extraction and matching, distance measurement and taking corresponding measures.
Description
Technical field
The present invention relates to a kind of forewarn system of vehicle going at night, more specifically, relate to composition and its implementation based on the forewarn system of the vehicle going at night of the far infrared image of monocular vision.
Background technology
Along with the development of auto-industry, there is increasing automobile to reach highway, in night or greasy weather during running automobile, although there is car light to illuminate road for you, danger still lies dormant in the dark forwardly.According to statistics, although night travel accounts for 1/4th in whole highway communication, the fatal accident that occurs during this has accounted for half, and the accident that night, bad visibility caused accounted for wherein 70%.So in the urgent need to developing a kind of forewarn system of car night running, to reduce accident rate.
Summary of the invention
The present invention proposes design and implementation method based on the vehicle going at night forewarn system of the far infrared image of monocular vision, the method can make automobile when night running in the time may bumping reminding driver or directly take brake measure.
In order to realize above-mentioned purpose, the present invention takes following scheme.System of the present invention composition mainly contains hardware and software two parts, and wherein hardware is following a few part: far infrared image collecting device, memory storage, treater, read out instrument, warning device, brake equipment; Software section has image processing program, control program.
The thermal infrared imager that far infrared image collecting device in the described hardware components adopts, it belongs to pick up camera, and it is used for gathering the real-time imaging of automobile working direction.
Memory storage in the described hardware components is the image that the memory image harvester gathers, and also is used for the result of memory image handler and the real-time data record of system's operation.
Treater in the described hardware components is mainly used to control program is processed and moved to the image file that reads from memory storage.
Read out instrument in the described hardware components is the place ahead real-time condition and the picture after the processing and the data after the section processes that show collection.
Warning device in the described hardware components is used for chaufeur or operator are reported to the police.
Brake equipment in the described hardware components is to take under special circumstances automatically automobile to be taked brake operating.
Image processing program in the described software section is that the image that reads from memory storage is carried out pretreatment and extracts target and find range.
The control program in the described software section is that given safety distance is compared with the distance that calculates in real time and differentiates automobile whether thereby safety is made control decision.
The present invention also provides a kind of vehicle going at night method for early warning of the far infrared image based on monocular vision, comprises the steps:
1) from memory device, reads the picture frame of the vehicle front that two width of cloth harvesters gather with the time gap of t second;
2) two width of cloth images that read are carried out pretreatment and extract interested part (ROI);
3) to 2) in ROI ask for respectively its histograms of oriented gradients (HOG);
4) 3) the basis on carry out identification and classification;
5) image is processed extracted objective body, whether the objective body of differentiating in two width of cloth images after the coupling is the unified goal body, if not words then return 1), if words then enter step 6);
6) by two width of cloth image calculation that extract objective body go out automobile from objects in front apart from d and with given safety distance d
iCompare, if d
i＜d then returns 1), if d
iD, warning then provided.By with the d of different stage
iCompare and take different type of alarms, will take automatic brake arrangement when reaching when highest level is warned.
Further, described step 2) specifically comprise following several step:
21) at first image having been carried out the Otsu method cuts apart;
22) then just fixed by carry out ROI based on the method for direction projection;
Further, described step 3) specifically comprises following several step:
31) normalized image;
32) compute gradient;
33) histogram of gradients is carried out the projection of regulation weight for each cell piece;
34) for cell degree of the comparing normalization method in each overlapping block piece;
35) histogram vectors in all block is combined into a large HOG proper vector together;
Further, described step 4) is specially:
What the discriminant classification of target used among the present invention is SVMs (SVM) algorithm.The human body candidate region positions in to infrared image, and after asking for its gradient orientation histogram, gradient orientation histogram can be classified to the target in the candidate region by the categorised decision function as input vector.Because gradient orientation histogram is high dimension vector, generally for multi-C vector at first dimensionality reduction improve arithmetic speed, use in the present invention SVMs (SVM) algorithm, just omitted this step.
Support vector machine method is that the VC that is based upon Statistical Learning Theory ties up on theoretical and the structure risk minimum principle basis, according to limited sample information in the complexity of model (namely to the study precision of specific training sample, Accuracy) and between the learning ability (namely identifying error-free the ability of arbitrary sample) seek optimal compromise, in the hope of obtaining best Generalization Ability (Generalization Ability).
Have 2 class images in the image library, be denoted as T={P1, P2, and the segregator { C1, C2 } of 2 semantemes is arranged.For each SVM segregator, its training set T={ (x1, y1), (x2, y2),,, (xn, yn) }, wherein (x1, y1) is the in advance given sample image through mark, xi ∈ R2 wherein, the HOG vector of its presentation graphs picture; { 1,1} is human body for+1 presentation graphs looks like to comprise this semanteme to yi ∈, and it is the position human body that-1 presentation graphs picture does not comprise this semanteme; Use SVM to train these sample images, obtain the semantic classifiers of image.Then can utilize these semantic classifiers to remove to differentiate the image that those do not mark.Get multiple image as exercise library by previous experiments.
Further, described step 5) specifically comprises as follows:
51) target is extracted;
52) object matching;
Further, described step 6) specifically comprises as follows:
Going out automobile by two width of cloth image calculation that extract objective body from the principle apart from d of object is: objects in front along with the automobile distance further or away from, on the image of taking objects in front in infrared thermography shared pixel also can along with camera further or away from and elongated or shorten, by in the computed image certain fixedly the variation of parameter shared pixel in image can calculate the distance that objects in front moves with respect to camera, and then the spacing of car before and after calculating.
Before and after wherein P1 and P2 are respectively in the target object of input picture certain fixedly parameter move in image shared pixel value, L is the distance of zoom.
Further, described step 22) be specially:
The ROI that the present invention takes just fixed method is: 1. level attitude determines; 2. the upright position determines.
For a width of cloth infrared image, define its brightness vertical projection curve for brightness in the corresponding image column greater than the number of the pixel of above-mentioned segmentation threshold.Because human body regional luminance of living in is higher, the shape that presents at the vertical projection curve, therefore, curve rising point that can be by determining drop shadow curve's protrusions part and curve drop point be divided into a series of projection band with curve and not comprise the level band of luminance pixel, and the projection band is corresponding may exist the zone of human body.Based on the These characteristics of brightness projection curve, can level be a series of belt-like zone with image segmentation, these district inclusions may have the candidate region of human body, the process of cutting apart is specific as follows:
1) adaptive selection Intensity segmentation threshold value records the number of pixels that each row meets brightness requirement at the brightness projection curve.
2) initial point (rising point) and the end point (drop point) of automatically search projection band.
3) with image segmentation for the brightness projection curve in projection with corresponding ribbon zone.Be the zone that possible have human body in these ribbon zones, by above-mentioned cut apart well will have human body region disconnecting out.Infrared image is being carried out after level cuts apart, can obtain the level attitude of human body candidate region, using the same method and the red ribbon image-region that obtains is carried out vertical segmentation determine its upright position.
Further, described step 31)-35) be specially:
What among the present invention the Feature Descriptor of the object detection of image is taked is histograms of oriented gradients.
Through behind the direction projection, the position of human body candidate region and size are known in the infrared image, and the approximate impact that is not subjected to illumination of human body image in infrared image, therefore adopt the histograms of oriented gradients solution strategies of single yardstick, single area in this programme, avoided complicated multiple dimensioned search procedure.Because distributing, the gradient of human body and non-human body with direction gradient certain difference is arranged, so can or divide a human body to identify with human body and background.Its detailed process is as follows:
ROI is represented as above with a rectangular window, ask for horizontal direction gradient and the vertical gradient of all pixels in the window, and then ask for the direction of gradient, then obtain this regional histograms of oriented gradients by the progression corresponding in the histogram that is divided into that gradient direction is discrete, because the symmetry of direction, when making up direction histogram, adopted the direction scope of [0 °, 180 °].Obvious, what we paid close attention in practice is the gradient direction of human body marginal portion in the image, the statistical property of these gradient directions has characterized the shape of human body, the intensity that a distinguishing feature of human body edge gradient is gradient will be higher than non-fringe region, therefore can gradient intensity as weights gradient direction be weighted the impact of body shape being expressed to weaken non-human body fringe region.Be in m * n rectangular window in size, order is positioned at (i, the coefficient of weight w (i, j) of pixel j) is the ratio of greatest gradient intensity in the gradient intensity at this some place and the candidate region, and the histogram of gradients H that then comprises zone within it can be expressed as:
Here k=1....z is histogrammic progression, h (i, j), v (i, j), g (i, j) and o (i, j) are respectively and are positioned at (i, the horizontal gradient of pixel j), vertical gradient, gradient intensity and gradient direction, δ are Kronecker delta function.
After asking for above-mentioned histogram, need to determine the used operator of histogram normalization method that so that the comparison that quantizes, common normalized operator has 1 norm, 2 norms and 3 norms etc. between different histograms; The parameter that another one need to be determined is histogrammic progression, and the progression that histogram comprises causes little direction rotation too responsive too much, and progression then causes structure too coarse very little, therefore need to select according to actual requirement.We will come the selection of above-mentioned two parameters to determine by experiment.
Further, described step 51) be specially:
Divided (figure) on the basis of class to extract from the nearest people of car as target in the above because image is two-dimentional, when three-dimensional scenic is mapped to two dimensional image from car nearest in two dimensional image be below.We can judge which human body is as from nearest namely our the required target of car according to the Y value of rectangle frame lower sideline on this basis.Determine the advantage of target the most nearby at this:
1) in follow-up processing, only processes for this target;
2) saved the time that follow-up many human body targets mate;
3) in the end do not need each target is carried out finding the solution of distance.
Further, described step 52) be specially:
After having extracted, target must mate the target that the front and back frame extracts, for follow-up range finding is prepared.The purpose of coupling is to guarantee that the target of extracting is same human body.The NMI(Normalized Moment of Inertia of image) invariant feature, be the normalization method rotor inertia of image, it has anti-tonal distortion and geometric distortion (such as translation, rotation, convergent-divergent) is had preferably and maintains the invariance, and the coupling accuracy is high, calculated amount is little, and speed is fast.
To width of cloth bianry image a: f (i, j), i=0,1, M 1; J=0,1, N 1, can be regarded as M * N particle on the XOY plane, and the gray scale of picture dot is exactly the quality of corresponding particle.Bond related notion of science is defined as follows image:
The pixel value sum is defined as the quality of this bianry image in the bianry image, is designated as m (f (i, j)), then:
The bianry image center of gravity is designated as CG (i, j), wherein:
Bianry image centers on wherein any point (i
0, j
0) rotor inertia be designated as J(i
0, j
0), then:
Bianry image normalization method rotor inertia is designated as NMI (f (i, j)), then:
In the formula, Π is that the bianry image intermediate value is the zone of " 1 ".Can find out that NMI is that bianry image is around the rotor inertia extraction of square root and its mass ratio of its center of gravity.
This forewarn system and method for early warning, by being installed in the thermal infrared imager on the automobile, gather in real time the video image of the road conditions of vehicle front, through the image that gathers being processed and being judged whether the place ahead has object, if the object that has is obtained the distance of object and work stall by corresponding distance-finding method, and make comparisons with the distance of given different stage, make respectively corresponding alerting signal prompting chaufeur or operator and make corresponding processing.Also do not make corresponding processing when highest level is reported to the police then take automatic brake measure if reach.Corresponding data when showing in real time the place ahead object view and routine processes by telltale simultaneously.The present invention is based on Design of warning and the implementation method of vehicle going at night of far infrared image of the monocular vision of infrared thermoviewer, its hardware configuration and respective algorithms are simple, be easy to realize, and with low cost, simple in structure, the advantage such as volume is little is fit to popularization and application the most, and the 26S Proteasome Structure and Function of vision sensor and human eye are similar, and the message form that obtains is also the most near human eye.
Description of drawings
The present invention is further described below in conjunction with drawings and Examples.
Fig. 1 is based on the constructional drawing of forewarn system of vehicle going at night of the far infrared image of monocular vision.
Fig. 2 is based on the implementation method diagram of circuit of forewarn system of vehicle going at night of the far infrared image of monocular vision.
1. the place ahead environment among the figure, 2. far infrared image collecting device, 3. memory storage, 4. automotive system, 5. treater, 6. image processing program, 7. control program, 8. read out instrument, 9. warning device, 10. automatic brake arrangement.
The specific embodiment
Below with reference to accompanying drawing, the preferred embodiments of the present invention are described in detail.
The connection mode of various devices is referring to Fig. 1, the forewarn system based on the vehicle going at night of the far infrared image of monocular vision of this preferred embodiment, the reason such as illumination deficiency during owing to vehicle going at night, normal camera head can not meet the demands, so need a kind of far infrared camera shot device 1, what choose in the present invention is thermal infrared imager, and it is installed in the Automobile head, be used for gathering the video information of automobile working direction, and be stored in the memory storage 2.Treater 3 reads different time sections from memory storage 2 frame carries out the processing of image processing program 6 and makes corresponding different decision-making by the processing of control program 7.Automotive system 4 is the operational factor of automobile (speed, acceleration/accel) to be transferred to treater carry out computing.
The concrete steps of implementation method are seen Fig. 2:
(1) from memory device 2, takes out two frames as processing image with t interval second;
(2) two width of cloth images that read are carried out pretreatment and extract interested part (ROI):
21) at first image having been carried out the Otsu method cuts apart;
22) then just fixed by carry out ROI based on the method for direction projection;
Its concrete steps are: 1. level attitude determines; 2. the upright position determines.
For a width of cloth infrared image, define its brightness vertical projection curve for brightness in the corresponding image column greater than the number of the pixel of above-mentioned segmentation threshold.Because human body regional luminance of living in is higher, the shape that presents at the vertical projection curve, therefore, curve rising point that can be by determining drop shadow curve's protrusions part and curve drop point be divided into a series of projection band with curve and not comprise the level band of luminance pixel, and the projection band is corresponding may exist the zone of human body.Based on the These characteristics of brightness projection curve, can level be a series of belt-like zone with image segmentation, these district inclusions may have the candidate region of human body, the process of cutting apart is specific as follows:
1) adaptive selection Intensity segmentation threshold value records the number of pixels that each row meets brightness requirement at the brightness projection curve.
2) initial point (rising point) and the end point (drop point) of automatically search projection band.
3) with image segmentation for the brightness projection curve in projection with corresponding ribbon zone.Be the zone that possible have human body in these ribbon zones, by above-mentioned cut apart well will have human body region disconnecting out.Infrared image is being carried out after level cuts apart, can obtain the level attitude of human body candidate region, using the same method and the red ribbon image-region that obtains is carried out vertical segmentation determine its upright position.
(3) ROI in (2) is asked for respectively its histograms of oriented gradients (HOG):
31) normalized image;
32) compute gradient;
33) histogram of gradients is carried out the projection of regulation weight for each cell piece;
34) for cell degree of the comparing normalization method in each overlapping block piece;
35) histogram vectors in all block is combined into a large HOG proper vector together;
Its detailed process is as follows:
ROI is represented as above with a rectangular window, ask for horizontal direction gradient and the vertical gradient of all pixels in the window, and then ask for the direction of gradient, then obtain this regional histograms of oriented gradients by the progression corresponding in the histogram that is divided into that gradient direction is discrete, because the symmetry of direction, when making up direction histogram, adopted the direction scope of [0 °, 180 °].Obvious, what we paid close attention in practice is the gradient direction of human body marginal portion in the image, the statistical property of these gradient directions has characterized the shape of human body, the intensity that a distinguishing feature of human body edge gradient is gradient will be higher than non-fringe region, therefore can gradient intensity as weights gradient direction be weighted the impact of body shape being expressed to weaken non-human body fringe region.Be in m * n rectangular window in size, order is positioned at (i, the coefficient of weight w (i, j) of pixel j) is the ratio of greatest gradient intensity in the gradient intensity at this some place and the candidate region, and the histogram of gradients H that then comprises zone within it can be expressed as:
Here k=1....z is histogrammic progression, h (i, j), v (i, j), g (i, j) and o (i, j) are respectively and are positioned at (i, the horizontal gradient of pixel j), vertical gradient, gradient intensity and gradient direction, δ are Kronecker delta function.
After asking for above-mentioned histogram, need to determine the used operator of histogram normalization method that so that the comparison that quantizes, common normalized operator has 1 norm, 2 norms and 3 norms etc. between different histograms; The parameter that another one need to be determined is histogrammic progression, and the progression that histogram comprises causes little direction rotation too responsive too much, and progression then causes structure too coarse very little, therefore need to select according to actual requirement.We will come the selection of above-mentioned two parameters to determine by experiment.
(4) on the basis of (3), carry out identification and classification:
The human body candidate region positions in to infrared image, and after asking for its gradient orientation histogram, gradient orientation histogram can be classified to the target in the candidate region by the categorised decision function as input vector.Because gradient orientation histogram is high dimension vector, generally for multi-C vector at first dimensionality reduction improve arithmetic speed, in this programme, use SVMs (SVM) algorithm, just omitted this step.
Support vector machine method is that the VC that is based upon Statistical Learning Theory ties up on theoretical and the structure risk minimum principle basis, according to limited sample information in the complexity of model (namely to the study precision of specific training sample, Accuracy) and between the learning ability (namely identifying error-free the ability of arbitrary sample) seek optimal compromise, in the hope of obtaining best Generalization Ability (Generalization Ability).
We have obtained the HOG vector that the β * ζ of every width of cloth image * η data form in (3), and wherein β is direction unit (bin) number among each cell, and ζ, η are respectively the number of cell among the number of block and the block.
Have 2 class images in the image library, be denoted as T={P1, P2, and the segregator { C1, C2 } of 2 semantemes is arranged.For each SVM segregator, its training set T={ (x1, y1), (x2, y2),,, (xn, yn) }, wherein (x1, y1) is the in advance given sample image through mark, xi ∈ R2 wherein, the HOG vector of its presentation graphs picture; { 1,1} is human body for+1 presentation graphs looks like to comprise this semanteme to yi ∈, and it is the position human body that-1 presentation graphs picture does not comprise this semanteme; Use SVM to train these sample images, obtain the semantic classifiers of image.Then can utilize these semantic classifiers to remove to differentiate the image that those do not mark.Get multiple image as exercise library by previous experiments.
(5) image is processed extracted objective body, whether the objective body of differentiating in two width of cloth images after the coupling is the unified goal body, if not words then return (1), if words then enter step (6):
51) target is extracted;
52) object matching;
Its detailed process is as follows:
Divided (figure) on the basis of class to extract from the nearest people of car as target in the above because image is two-dimentional, when three-dimensional scenic is mapped to two dimensional image from car nearest in two dimensional image be below.We can judge which human body is as from nearest namely our the required target of car according to the Y value of rectangle frame lower sideline on this basis.
After having extracted, target must mate the target that the front and back frame extracts, for follow-up range finding is prepared.The purpose of coupling is to guarantee that the target of extracting is same human body.The NMI(Normalized Moment of Inertia of image) invariant feature, be the normalization method rotor inertia of image, it has anti-tonal distortion and geometric distortion (such as translation, rotation, convergent-divergent) is had preferably and maintains the invariance, and the coupling accuracy is high, calculated amount is little, and speed is fast.
To width of cloth bianry image a: f (i, j), i=0,1, M 1; J=0,1, N 1, can be regarded as M * N particle on the XOY plane, and the gray scale of picture dot is exactly the quality of corresponding particle.Bond related notion of science is defined as follows image:
The pixel value sum is defined as the quality of this bianry image in the bianry image, is designated as m (f (i, j)), then:
The bianry image center of gravity is designated as CG (i, j), wherein:
Bianry image centers on wherein any point (i
0, j
0) rotor inertia be designated as J(i
0, j
0), then:
Bianry image normalization method rotor inertia is designated as NMI (f (i, j)), then:
In the formula, Π is that the bianry image intermediate value is the zone of " 1 ".Can find out that NMI is that bianry image is around the rotor inertia extraction of square root and its mass ratio of its center of gravity.
(6) mated after, carry out be the range finding.Going out automobile by two width of cloth image calculation that extract objective body in the present invention from the principle apart from d of object is: objects in front along with the automobile distance further or away from, on the image of taking objects in front in infrared thermography shared pixel also can along with camera further or away from and elongated or shorten, by in the computed image certain fixedly the variation of parameter shared pixel in image can calculate the distance that objects in front moves with respect to camera, and then the spacing of car before and after calculating.
Before and after wherein P1 and P2 are respectively in the target object of input picture certain fixedly parameter move in image shared pixel value, L is the distance of zoom.The height of the objective body in two width of cloth bianry images or width be along with can changing to some extent with the distance of automobile shared pixel count in image, so what choose in the present invention is the height of image.Be that P1 and P2 are respectively t and obtain objective body among the two frame figure around the second.L is the namely distance of moving vehicle of the distance that moves of infrared thermoviewer, can obtain by the data that automotive system provides.
(7) be exactly at last the safety distance d apart from several different stages given in d and the memory device by measuring
iCompare, if d
i＜d then returns (1), if d
iD, warning then provided.By with the d of different stage
iCompare and take different type of alarms, will take automatic brake arrangement when reaching when highest level is warned.3 other d of level have been set among the present invention
i, be respectively elementary d
1(the object flicker on short ring and the telltale), intermediate d
2(the object flicker on length ring interruption and the telltale), senior d
3(yowl and take the autobrake measure after two seconds).
Claims (10)
1. based on vehicle going at night forewarn system and the method for far infrared image, it is characterized in that: the composition of system hardware and software and the step of realization.
2. hardware according to claim 1 forms, and it is characterized in that: be by the far infrared image collecting device, and memory storage, treater, read out instrument, warning device, brake equipment forms.
3. software according to claim 1 forms, and it is characterized in that: be comprised of image processing program and control program.
4. far infrared image collecting device according to claim 2 is characterized in that: be for the video that gathers the automobile working direction and be stored in the memory storage of back, usefulness is thermal infrared imager in the present invention, and it is installed in or headstock anterior.
5. described according to claim 2, warning device is characterized in that: employing be that ring device and telltale are beated, and according to different grades the different modes that pipes is arranged, set 3 other d of level among the present invention
i, be respectively elementary d
1(the object flicker on short ring and the telltale), intermediate d
2(the object flicker on length ring interruption and the telltale), senior d
3(yowl and take the autobrake measure after two seconds).
6. described according to claim 3, its image processing program is characterized in that: be just fixed by ROI, and histograms of oriented gradients, identification and classification, target is extracted with coupling and is formed.
7. described according to claim 6, the method that each step adopts is characterized in that: it is first image to be carried out the method that the Otsu method cuts apart based on direction projection again to obtain that ROI extracts, the Feature Descriptor that the present invention takes is histograms of oriented gradients (HOG), what identification and classification was taked among the present invention is SVMs (SVM) algorithm, and the proper vector of extracting among the present invention is that the NMI of image is indeformable.
8. described according to claim 7, histograms of oriented gradients (HOG) is characterised in that: the histograms of oriented gradients solution strategies that has adopted single yardstick, single area, avoided complicated multiple dimensioned search procedure, because distributing, the gradient of human body and non-human body with direction gradient certain difference is arranged, so can or divide a human body to identify with human body and background, histogrammic progression and normalized operator be to come by experiment to determine.
9. described according to claim 7, identification and classification is SVMs (SVM) algorithm of taking, and it is characterized in that: have 2 class images in the image library, be denoted as T={P1, P2 }, and the segregator { C1, C2 } of 2 semantemes is arranged, for each SVM segregator, its training set T={ (x1, y1), (x2, y2), (xn, yn) }, (x1 wherein, y1) be the in advance given sample image through mark, xi ∈ R2 wherein, the HOG vector of its presentation graphs picture; { 1,1} is human body for+1 presentation graphs looks like to comprise this semanteme to yi ∈, and it is the position human body that-1 presentation graphs picture does not comprise this semanteme; Use SVM to train these sample images, obtain the semantic classifiers of image, then can utilize these semantic classifiers to remove to differentiate the image that those do not mark.
10. described according to claim 7, feature extraction and coupling are characterised in that: will be from the nearest people of car as target among the present invention, the NMI invariant feature of image, it has anti-tonal distortion and geometric distortion (such as translation, rotation, convergent-divergent) is had preferably and maintains the invariance, and the coupling accuracy is high, calculated amount is little, and speed is fast.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2012104548976A CN102975659A (en) | 2012-11-14 | 2012-11-14 | Automobile night driving pre-warning system and method based on far infrared image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2012104548976A CN102975659A (en) | 2012-11-14 | 2012-11-14 | Automobile night driving pre-warning system and method based on far infrared image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN102975659A true CN102975659A (en) | 2013-03-20 |
Family
ID=47850117
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2012104548976A Pending CN102975659A (en) | 2012-11-14 | 2012-11-14 | Automobile night driving pre-warning system and method based on far infrared image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102975659A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104290662A (en) * | 2014-09-30 | 2015-01-21 | 成都市晶林科技有限公司 | Driving assisting system and driving assisting method |
| CN104442571A (en) * | 2014-11-26 | 2015-03-25 | 重庆长安汽车股份有限公司 | Night vision navigation integration system and control method |
| CN104463902A (en) * | 2013-09-25 | 2015-03-25 | 北京环境特性研究所 | Stationary target elimination method based on NMI feature |
| CN105799574A (en) * | 2016-05-31 | 2016-07-27 | 常州星宇车灯股份有限公司 | Intelligent night driver-assistance system |
| CN107038711A (en) * | 2016-02-02 | 2017-08-11 | 财团法人资讯工业策进会 | Adaptive evolution type vehicle lamp signal detection tracking and identification system and method |
| CN107292246A (en) * | 2017-06-05 | 2017-10-24 | 河海大学 | Infrared human body target identification method based on HOG PCA and transfer learning |
| CN108495030A (en) * | 2018-03-16 | 2018-09-04 | 维沃移动通信有限公司 | A kind of image processing method and mobile terminal |
| CN108883725A (en) * | 2016-03-06 | 2018-11-23 | 预见汽车有限责任公司 | A kind of driving vehicle warning system and method |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN201703375U (en) * | 2010-05-26 | 2011-01-12 | 浙江机电职业技术学院 | Anti-collision system of automobile |
| CN102395009A (en) * | 2011-06-29 | 2012-03-28 | 广东好帮手电子科技股份有限公司 | Vehicle-mounted infrared night vision system with pedestrian identification function |
| CN102765365A (en) * | 2011-05-06 | 2012-11-07 | 香港生产力促进局 | Pedestrian detection method based on machine vision and pedestrian anti-collision warning system based on machine vision |
-
2012
- 2012-11-14 CN CN2012104548976A patent/CN102975659A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN201703375U (en) * | 2010-05-26 | 2011-01-12 | 浙江机电职业技术学院 | Anti-collision system of automobile |
| CN102765365A (en) * | 2011-05-06 | 2012-11-07 | 香港生产力促进局 | Pedestrian detection method based on machine vision and pedestrian anti-collision warning system based on machine vision |
| CN102395009A (en) * | 2011-06-29 | 2012-03-28 | 广东好帮手电子科技股份有限公司 | Vehicle-mounted infrared night vision system with pedestrian identification function |
Non-Patent Citations (2)
| Title |
|---|
| 李建福: "红外图像中人体目标检测、跟踪及其行为", 《万方学位论文》 * |
| 顾柏园: "基于单目视觉的安全车距预警系统研究", 《中国博士学位论文全文数据库》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104463902A (en) * | 2013-09-25 | 2015-03-25 | 北京环境特性研究所 | Stationary target elimination method based on NMI feature |
| CN104463902B (en) * | 2013-09-25 | 2017-12-22 | 北京环境特性研究所 | A kind of static target removing method based on NMI features |
| CN104290662A (en) * | 2014-09-30 | 2015-01-21 | 成都市晶林科技有限公司 | Driving assisting system and driving assisting method |
| CN104442571A (en) * | 2014-11-26 | 2015-03-25 | 重庆长安汽车股份有限公司 | Night vision navigation integration system and control method |
| CN107038711A (en) * | 2016-02-02 | 2017-08-11 | 财团法人资讯工业策进会 | Adaptive evolution type vehicle lamp signal detection tracking and identification system and method |
| CN107038711B (en) * | 2016-02-02 | 2020-08-21 | 财团法人资讯工业策进会 | Vehicle light signal detecting, tracking and identifying system and method |
| CN108883725A (en) * | 2016-03-06 | 2018-11-23 | 预见汽车有限责任公司 | A kind of driving vehicle warning system and method |
| CN108883725B (en) * | 2016-03-06 | 2021-02-23 | 预见汽车有限责任公司 | Driving vehicle alarm system and method |
| CN105799574A (en) * | 2016-05-31 | 2016-07-27 | 常州星宇车灯股份有限公司 | Intelligent night driver-assistance system |
| CN105799574B (en) * | 2016-05-31 | 2018-08-21 | 常州星宇车灯股份有限公司 | Night intelligence DAS (Driver Assistant System) |
| CN107292246A (en) * | 2017-06-05 | 2017-10-24 | 河海大学 | Infrared human body target identification method based on HOG PCA and transfer learning |
| CN108495030A (en) * | 2018-03-16 | 2018-09-04 | 维沃移动通信有限公司 | A kind of image processing method and mobile terminal |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102975659A (en) | Automobile night driving pre-warning system and method based on far infrared image | |
| Possatti et al. | Traffic light recognition using deep learning and prior maps for autonomous cars | |
| CN100462047C (en) | Safe driving auxiliary device based on omnidirectional computer vision | |
| CN102867417B (en) | Taxi anti-forgery system and taxi anti-forgery method | |
| JP5591779B2 (en) | Road use vulnerable person protection system | |
| CN102073851B (en) | Method and system for automatically identifying urban traffic accident | |
| US9070023B2 (en) | System and method of alerting a driver that visual perception of pedestrian may be difficult | |
| CN112700470A (en) | Target detection and track extraction method based on traffic video stream | |
| CN110379168B (en) | Traffic vehicle information acquisition method based on Mask R-CNN | |
| CN105404857A (en) | Infrared-based night intelligent vehicle front pedestrian detection method | |
| EP2813973B1 (en) | Method and system for processing video image | |
| CN112825192B (en) | Object identification system and method based on machine learning | |
| CN104200668A (en) | Image-analysis-based detection method for helmet-free motorcycle driving violation event | |
| CN103310206B (en) | A kind of vehicle using motor detection method based on many features and multiframe information fusion | |
| Romera et al. | A Real-Time Multi-scale Vehicle Detection and Tracking Approach for Smartphones. | |
| CN104378539A (en) | Scene-adaptive video structuring semantic extraction camera and method thereof | |
| JP2012221162A (en) | Object detection device and program | |
| CN114818819A (en) | Road obstacle detection method based on millimeter wave radar and visual signal | |
| Xiao et al. | Detection of drivers visual attention using smartphone | |
| Yun et al. | Video-based detection and analysis of driver distraction and inattention | |
| JP2004086417A (en) | Method and device for detecting pedestrian on zebra crossing | |
| Philipsen et al. | Day and night-time drive analysis using stereo vision for naturalistic driving studies | |
| CN116012822B (en) | Fatigue driving identification method and device and electronic equipment | |
| Álvarez et al. | Perception advances in outdoor vehicle detection for automatic cruise control | |
| Ponte et al. | Using specialised cyclist detection software to count cyclists and determine cyclist travel speed from video |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20130320 |