CN101577052B - Device and method for detecting vehicles by overlooking - Google Patents
Device and method for detecting vehicles by overlooking Download PDFInfo
- Publication number
- CN101577052B CN101577052B CN2009101429367A CN200910142936A CN101577052B CN 101577052 B CN101577052 B CN 101577052B CN 2009101429367 A CN2009101429367 A CN 2009101429367A CN 200910142936 A CN200910142936 A CN 200910142936A CN 101577052 B CN101577052 B CN 101577052B
- Authority
- CN
- China
- Prior art keywords
- vehicle
- candidate region
- frame
- image
- zone
- 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.)
- Expired - Fee Related
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
A device for detecting vehicles by overlooking comprises a translation unit, a statistical unit, a calculation unit and a classification unit; wherein, the translation unit is used for translating a first frame vehicle image and an adjacent second frame vehicle image thereof to coincide in a world coordinate system; the statistical unit is used for respectively obtaining the candidate areas of thevehicles in the first frame vehicle image and the second frame vehicle image; the calculation unit is used for carrying out image difference on the candidate areas of the vehicles and pixel areas cor respondingly coinciding with the candidate areas to obtain the candidate areas of the moving vehicles in the second frame vehicle image relative to the first frame vehicle image; the classification unit is used for classifying the candidate areas of the moving vehicles according to a classifier obtained through specimen training to detect the areas of the moving vehicles in the second frame vehicle image relative to the first frame vehicle image. The device has high detection rate and low misstatement rate under the premise of ensuring the requirement of real-time.
Description
Technical field
The invention belongs to intelligent transportation field, relate in particular to the apparatus and method that a kind of vehicles by overlooking detects.
Background technology
City Traffic Monitor System is as reducing traffic hazard and congested in traffic effective technology, in each big city widespread use.The hollow panel of bowing that utilizes unmanned spacecraft lift-launch camera has become a research focus because its cost is low, the advantage of looking away.Just because of this, under the hollow panel of bowing, detect moving vehicle in the road traffic and become the gordian technique that research circle and industrial community are very paid close attention to.
In recent years, the researcher has obtained many achievements aspect vehicle detection both at home and abroad, and the algorithm of for this reason designing must satisfy following requirement: 1) real-time demand: the travelling speed of algorithm must be faster than the shooting speed of video; 2) verification and measurement ratio demand: much more as far as possible algorithm must must detect the moving vehicles in the video flowing; 3) rate of false alarm demand: algorithm must reduce non-moving vehicle as far as possible and be reported by mistake and be moving vehicle;
Existing vehicle detecting algorithm mainly can be divided three classes and exist following shortcoming: 1) frame difference method: it is very big influenced by illumination and environmental change; 2) background subtraction division: can not use and motion platform; 3) optical flow method: calculation of complex, can not satisfy the real-time demand.Be necessary to design a kind of effective vehicle detection technology for this reason.
Summary of the invention
Purpose of the present invention is intended to one of solve the aforementioned problems in the prior at least.
For this reason, embodiments of the invention propose a kind of vehicles by overlooking detection method and device with high detection rate and low rate of false alarm.
According to an aspect of the present invention, the vehicles by overlooking detection method of the embodiment of the invention may further comprise the steps: a) the second frame vehicle image that is adjacent of the translation first frame vehicle image overlaps on world coordinate system; B) utilize pixel color to add up the vehicle candidate region that obtains respectively in described first frame vehicle image and the described second frame vehicle image; C) image difference calculating is carried out in described vehicle candidate region and its corresponding pixel region that overlaps, to obtain the moving vehicle candidate region of the described relatively first frame vehicle image of the described second frame vehicle image; And d) in the sorter of sample training gained, classified in described moving vehicle candidate region, to detect the moving vehicle zone of the described relatively first frame vehicle image of the described second frame vehicle image.
The further embodiment according to the present invention, described step b may further comprise the steps: the b1) ratio of various pixel color in described first frame vehicle image of statistics and the described second frame vehicle image; B2) according to the definite respectively described first frame vehicle image of pixel color of maximum ratio and the road area in the described second frame vehicle image; And b3) removes described road area and determine described vehicle candidate region.
The further embodiment according to the present invention also comprises and utilizes the vehicle dimension size to filter out the construction zone in the described vehicle candidate region and/or the step in noise zone.Pixel region area and the described vehicle candidate region by more described vehicle dimension correspondence wherein will be fallen much larger than the construction zone in described vehicle dimension zone and/or much smaller than the noise area filter in described vehicle dimension zone in the described vehicle candidate region.
The further embodiment according to the present invention also comprises and utilizes the vehicle dimension size to determine described vehicle candidate region.Wherein, the zone that is substantially equal to respective pixel zone, described vehicle dimension zone in the described vehicle candidate region is defined as described vehicle candidate region by the pixel region area and the described vehicle candidate region of more described vehicle dimension correspondence.
The further embodiment according to the present invention, described step a may further comprise the steps: a1) extract predetermined number of pixels zone in the described second frame vehicle image; A2) obtain corresponding minimum zone with the pixel difference in described predetermined number of pixels zone in the described first frame vehicle image; And a3) the described predetermined number of pixels of translation zone overlaps with described corresponding region.
The further embodiment according to the present invention, described steps d may further comprise the steps: the size of d1) adjusting described moving vehicle candidate region is big or small consistent with described training sample; D2) eigenwert of the moving vehicle candidate region of the described adjustment of calculating; D3) eigenwert and the described training sample characteristic threshold value of more described moving vehicle candidate region are to judge moving vehicle zone described in the described moving vehicle candidate region.
The further embodiment according to the present invention, described sorter are stacked sorter or tree classifier.
According to a further aspect in the invention, the vehicles by overlooking pick-up unit of the embodiment of the invention comprises: translation unit, and the second frame vehicle image that the described translation unit translation first frame vehicle image is adjacent overlaps on world coordinate system; Statistic unit, described statistic unit utilize pixel color to add up the vehicle candidate region that obtains respectively in described first frame vehicle image and the described second frame vehicle image; Computing unit, described computing unit carries out image difference calculating to described vehicle candidate region and its corresponding pixel region that overlaps, to obtain the moving vehicle candidate region of the described relatively first frame vehicle image of the described second frame vehicle image; And taxon, according to the sorter of sample training gained classified in described moving vehicle candidate region, to detect the moving vehicle zone of the described relatively first frame vehicle image of the described second frame vehicle image.
The further embodiment according to the present invention, described statistic unit is determined road area and described vehicle candidate region in described first frame vehicle image and the described second frame vehicle image according to the maximum ratio of various pixel color in described first frame vehicle image of adding up and the described second frame vehicle image.
The further embodiment according to the present invention also comprises filter element, and described filter element utilizes the vehicle dimension size to filter out construction zone and/or noise zone in the described vehicle candidate region.Perhaps comprise determining unit, described determining unit utilizes the vehicle dimension size to determine described vehicle candidate region.
The further embodiment according to the present invention, the size that described taxon is adjusted described moving vehicle candidate region is big or small consistent with described training sample; Calculate the eigenwert of the moving vehicle candidate region of described adjustment; And the characteristic threshold value size of the eigenwert of more described moving vehicle candidate region and described training sample, to judge moving vehicle zone described in the described moving vehicle candidate region.
The further embodiment according to the present invention, described taxon is stacked sorter or tree classifier.
The present invention proposes a kind of by thick vehicle checking method to essence, utilize the fast advantage of image processing speed from region-of-interest, to obtain the moving vehicle candidate region rapidly earlier, utilize the high advantage of sorting algorithm degree of accuracy then, from the candidate region, accurately obtain moving vehicle information.The moving vehicle that the present invention is directed under the hollow panel of bowing detects, and detects moving vehicle from complicated urban transportation background, under the prerequisite that guarantees the real-time demand, has high detection rate and low rate of false alarm.
Aspect that the present invention adds and advantage part in the following description provide, and part will become obviously from the following description, or recognize by practice of the present invention.
Description of drawings
Above-mentioned and/or additional aspect of the present invention and advantage are from obviously and easily understanding becoming the description of embodiment below in conjunction with accompanying drawing, wherein:
Fig. 1 is the block diagram of vehicles by overlooking pick-up unit of the present invention;
Fig. 2 is the flow chart of steps of vehicles by overlooking detection method of the present invention; And
Fig. 3 is the vehicle detection design sketch of the embodiment of the invention.
Embodiment
Describe embodiments of the invention below in detail, the example of described embodiment is shown in the drawings, and wherein identical from start to finish or similar label is represented identical or similar elements or the element with identical or similar functions.Below by the embodiment that is described with reference to the drawings is exemplary, only is used to explain the present invention, and can not be interpreted as limitation of the present invention.
With reference now to Fig. 1,, Fig. 1 is the block diagram of vehicles by overlooking pick-up unit of the present invention.As shown in the figure, vehicle detection apparatus of the present invention comprises translation unit 12, statistic unit 14, computing unit 16 and taxon 18.
Consider time interval between general video flowing two two field pictures be 20ms to 40ms, and during this period of time, the empty aircraft that detects of bowing can be regarded as basically and do rectilinear motion and without spin.Here, translation unit 12 can only be got several representative zones on second two field picture, and on first two field picture, seek zone with the second two field picture pixel difference minimum, and do translation according to these zones, two two field pictures are dropped in the unified world coordinate system substantially.
14 of statistic units utilize pixel color to add up the vehicle candidate region that obtains respectively in the first frame vehicle image and the second frame vehicle image.At an embodiment, consider that under the sky shooting height of bowing for example under 60 to 90 meters the height, the road color is the proportion maximum in the color histogram of entire image.That is, the widest color of area is considered to the road color in the image.
Therefore, statistic unit 14 is by the color histogram of statistical picture, and the color that can obtain the proportion maximum is the road color, thereby obtains road information.The road area color that utilization obtains is divided into disjunct zone with entire image, and the road area in first frame and second two field picture is got rid of, thereby can extract the vehicle candidate region.
In fact, may comprise in the vehicle candidate region that obtains that there are the zone of notable difference in vehicle and roadside buildings etc. and road color.Consider that the shared elemental area of the buildings in roadside in the image is far longer than the elemental area of vehicle correspondence, and the shared elemental area of some small noises is far smaller than the elemental area of vehicle, therefore can filter out some non-vehicle region with vehicle size
Therefore, in one embodiment, vehicle detection apparatus of the present invention can also comprise the filter element (not shown), construction zone and/or noise zone that filter element utilizes the size of vehicle reality to filter out may to exist in the above-mentioned vehicle candidate region.And/or, vehicle detection apparatus of the present invention can also comprise the determining unit (not shown), determining unit utilizes the vehicle dimension size to determine the vehicle candidate region.Thereby utilize filter element and/or determining unit can further improve the accuracy of the vehicle candidate region that obtains.
In one embodiment, suppose region area in the vehicle candidate region be p (pixel * pixel), if
Perhaps
Set up, represent that then this zone is not a vehicle region, wherein A represents the region area size that vehicle is actual shared, and S is an image analytic degree, and f is a focus of camera, and H is the height of video camera.
Filter element by vehicle dimension correspondence relatively the pixel region area and the pixel region in the top definite vehicle candidate region, will fall much larger than the construction zone in vehicle dimension zone and/or much smaller than the noise area filter of vehicle dimension region area in two two field pictures wherein.
Certainly, if basically
Then determining unit can determine further that this zone corresponds to the vehicle candidate region.
After above-mentioned definite vehicle candidate region, vehicle candidate region in the 16 pairs first frame vehicle images of computing unit and second vehicle image is carried out image difference respectively and is calculated, be the corresponding pixel region that overlaps of each vehicle candidate region of Difference Calculation and its, it is poor that the pixel value of respective pixel on the pixel value of each point on second two field picture and first two field picture is done.If the pixel difference be 0 presentation video do not change, if non-0 then presentation video at this some variation has taken place, promptly produced motion.Produce the candidate region of the zone of motion as moving vehicle by record, computing unit 16 can obtain the moving vehicle candidate region that the relative first frame vehicle image vehicle of the second frame vehicle image produces motion.
So far, the present invention at first utilizes the image processing operations of said units to obtain the moving vehicle candidate region rapidly from region-of-interest, therefore has fireballing advantage.But the moving vehicle candidate region that reality is obtained above also may comprise the non-vehicle object of some motions of roadside, for example pedestrian and motorcycle or the like, and therefore need carry out further refinement to the candidate region accurately obtains moving vehicle.
Below, on the above-mentioned basis that obtains the moving vehicle candidate region, taxon 18 is carried out precise classification according to the sorter that training sample obtains to these moving vehicle candidate regions, thereby detects the moving vehicle zone of the relative first frame vehicle image of the second frame vehicle image more accurately.
Taxon 18 can be the equipment that stacked sorter, tree classifier etc. are used for sample classification.
Comprise in the taxon 18 by positive sample (vehicle pictures) and negative sample (non-vehicle pictures) and train the training sample that obtains under off-line state, the quality and quantity of used sample will directly influence the classification results of sorter.At sorting phase at first, taxon 18 is read in the used feature and the eigenwert of characteristic set, the number of plies of sorter, each layer of the good sample of precondition.
Taxon 18 can be utilized bilinear transformation, and it is consistent with the training sample size that size is adjusted in all vehicle candidate regions that obtain, and calculate the eigenwert size of adjusted moving vehicle candidate region.Then, by the eigenwert of comparing motion vehicle candidate region and the characteristic threshold value size of training sample, thereby further from the moving vehicle candidate region, sort out the moving vehicle zone.
If taxon 18 judges relatively that by eigenwert corresponding moving vehicle candidate region is positive sample, then keep; Otherwise delete this candidate region.
In stacked sorter embodiment, for any one single classifier in the sorter, use the threshold test moving vehicle candidate region of each feature i wherein, obtain the comparative result f of character pair
i(x): 0 is expressed as non-vehicle, 1 expression vehicle.The weight of all T the feature i correspondences in calculated candidate zone f (the x)=∑ of suing for peace then
I=1toTw
if
i(x), if f (x) 〉=θ, wherein θ represents the sample judgment threshold, judges that then the candidate region is the moving vehicle zone.If be judged as vehicle, then enter down one deck single classifier, carry out corresponding this layer classification, otherwise should from the candidate region, delete in the zone.Until the classification of finishing all layers, the moving vehicle that then detects of the vehicle of Que Dinging at this moment for the present invention.
In addition, the invention allows for a kind of vehicles by overlooking detection method.The present invention includes following steps: a) the second frame vehicle image that is adjacent of the translation first frame vehicle image overlaps on world coordinate system; B) utilize pixel color to add up the vehicle candidate region that obtains respectively in the first frame vehicle image and the second frame vehicle image; C) image difference calculating is carried out in vehicle candidate region and its corresponding pixel region that overlaps, to obtain the moving vehicle candidate region of the described relatively first frame vehicle image of the second frame vehicle image; And d) in the sorter that sample training obtains, classified in the moving vehicle candidate region, to detect the moving vehicle zone of the relative first frame vehicle image of the second frame vehicle image.
Below with reference to Fig. 2, will describe the step of vehicles by overlooking detection method of the present invention in detail in conjunction with this embodiment.
As shown in the figure, from original input video, read adjacent two frame vehicle images, first two field picture and second two field picture.For the difference of two two field pictures before and after detecting in follow-up image difference, here adjacent can be two frames with 3-6 frame period.
In step 102, second two field picture and first two field picture are carried out translation, make them on world coordinate system, overlap.In one embodiment, can only get several representative zones on second two field picture, and on first two field picture, seek zone with the second two field picture pixel difference minimum, and do translation according to these zones, two two field pictures are dropped in the unified world coordinate system substantially.
Then, in step 104 and step 108, extract non-road area in first two field picture and second two field picture respectively.Consider that the road color is the proportion maximum in the color histogram of entire image under the sky shooting height of bowing.That is, the widest color of area is considered to the road color in the image.Therefore, by the color histogram of statistical picture, the color that can obtain the proportion maximum is the road color, thereby obtains road information.The road area color that utilization obtains is got rid of the road area in first frame and second two field picture, thereby can be extracted non-road area.
The non-road area here can be the rough vehicle candidate region of determining, in fact, may comprise in the vehicle candidate region that obtains that there are the zone of notable difference in vehicle and roadside buildings etc. and road color.Certainly in order further to improve the precision of obtaining of vehicle candidate region, can follow execution in step 106 and step 108 below, thereby the non-vehicle region in the zone of extracting in first two field picture and second two field picture is deleted.
In step 106 and step 108, consider that the shared elemental area of the buildings in roadside in the image is far longer than the elemental area of vehicle correspondence, and the shared elemental area of some small noises is far smaller than the elemental area of vehicle, therefore can delete wherein some non-vehicle region with vehicle size
In one embodiment, can utilize the size of vehicle reality to filter out the construction zone and/or the noise zone that may exist in the above-mentioned vehicle candidate region.In one embodiment, the present invention can also utilize the vehicle dimension size to determine the vehicle candidate region, promptly deletes non-vehicle region, thereby further improves the accuracy of the vehicle candidate region that obtains.
In one embodiment, suppose region area in the vehicle candidate region be p (pixel * pixel), if
Perhaps
Set up, represent that then this zone is not a vehicle region, wherein A represents the region area size that vehicle is actual shared, and S is an image analytic degree, and f is a focus of camera, and H is the height of video camera.
Pixel region in pixel region area by vehicle dimension correspondence relatively and the top definite vehicle candidate region will fall much larger than the construction zone in vehicle dimension zone and/or much smaller than the noise area filter of vehicle dimension region area in two two field pictures wherein.
Certainly, if basically
Can determine that then this zone corresponds to the vehicle candidate region.
After above-mentioned definite vehicle candidate region, in the step 112 the vehicle candidate region in the first frame vehicle image and second vehicle image being carried out image difference respectively calculates, be the corresponding pixel region that overlaps of each vehicle candidate region of Difference Calculation and its, it is poor that the pixel value of respective pixel on the pixel value of each point on second two field picture and first two field picture is done.If the pixel difference be 0 presentation video do not change, if non-0 then presentation video at this some variation has taken place, promptly produced motion.Produce the candidate region of the zone of motion by record, can obtain the moving vehicle candidate region that the relative first frame vehicle image vehicle of the second frame vehicle image produces motion as moving vehicle.
So far, the present invention utilizes above-mentioned image processing operations to obtain the moving vehicle candidate region rapidly from region-of-interest.In fact, the moving vehicle candidate region that obtains also may comprise the non-vehicle object of some motions of roadside, for example pedestrian and motorcycle or the like, and therefore need carry out further refinement to the candidate region accurately obtains moving vehicle.
Therefore then, on the above-mentioned basis that obtains the moving vehicle candidate region, in conjunction with the sample training sorter further precise classification is carried out in these moving vehicle candidate regions, thereby detect the moving vehicle zone of the relative first frame vehicle image of the second frame vehicle image more accurately.
Sorter can be the equipment that stacked sorter, tree classifier etc. are used for sample classification, comprising by positive sample (vehicle pictures) and negative sample (non-vehicle pictures).Under off-line state, obtain training classifier by step 202 training sample.Carrying out the branch time-like, at first utilize bilinear transformation, with all training sample that the adjustment of vehicle candidate region is big or small and sorter requires sizes consistent (step 114) that obtain.
Then, in sorter classification step 116, when beginning, sorting phase at first reads in the used feature and the eigenwert of characteristic set, the number of plies of sorter, each layer of the good sample of precondition.
By calculating the eigenwert size of adjusted moving vehicle candidate region, and the characteristic threshold value size of the eigenwert of comparing motion vehicle candidate region and training sample, thereby further from the moving vehicle candidate region, sort out the moving vehicle zone.
If judge that in classification step corresponding moving vehicle candidate region is positive sample, then keep, promptly export detected moving vehicle; Otherwise delete this candidate region.
In stacked sorter embodiment, for any one single classifier in the sorter, use the threshold test moving vehicle candidate region of each feature i wherein, obtain the comparative result f of character pair
i(x): 0 is expressed as non-vehicle, 1 expression vehicle.The weight of all T the feature i correspondences in calculated candidate zone f (the x)=∑ of suing for peace then
I=1toTw
if
i(x), if f (x) 〉=θ, wherein θ represents the sample judgment threshold, judges that then the candidate region is the moving vehicle zone.If be judged as vehicle, then enter down one deck single classifier, carry out corresponding this layer classification, otherwise should from the candidate region, delete in the zone.Until the classification of finishing all layers, the moving vehicle that then detects of the vehicle of Que Dinging at this moment for the present invention.
The moving vehicle that Fig. 3 has provided one embodiment of the invention detects design sketch, and the square frame among the figure partly is and utilizes the detected moving vehicle of the present invention zone.From figure as can be known, the present invention has higher detection rate and low rate of false alarm.
The present invention proposes a kind of by thick vehicle checking method to essence, utilize the fast advantage of image processing speed from region-of-interest, to obtain the moving vehicle candidate region rapidly earlier, utilize the high advantage of sorting algorithm degree of accuracy then, from the candidate region, accurately obtain moving vehicle information.The present invention detects at the moving vehicle under the hollow panel of bowing specially, provide under bow control video or the prerequisite the user by the unmanned spacecraft capture video, from complicated urban transportation background, detect moving vehicle, under the prerequisite that guarantees the real-time demand, have high detection rate and low rate of false alarm.Experiment shows that verification and measurement ratio of the present invention can reach 92%, and rate of false alarm is lower than 3%, and can satisfy the real-time demand.
Although illustrated and described embodiments of the invention, for the ordinary skill in the art, be appreciated that without departing from the principles and spirit of the present invention and can carry out multiple variation, modification, replacement and modification that scope of the present invention is by claims and be equal to and limit to these embodiment.
Claims (13)
1. a vehicles by overlooking detection method is characterized in that, may further comprise the steps:
A) the second frame vehicle image that is adjacent of the translation first frame vehicle image overlaps on world coordinate system;
B) utilize pixel color to add up the vehicle candidate region that obtains respectively in described first frame vehicle image and the described second frame vehicle image, described step b may further comprise the steps:
B1) ratio of various pixel color in described first frame vehicle image of statistics and the described second frame vehicle image;
B2) according to the definite respectively described first frame vehicle image of pixel color of maximum ratio and the road area in the described second frame vehicle image; And
B3) remove described road area and determine described vehicle candidate region;
C) image difference calculating is carried out in described vehicle candidate region and its corresponding pixel region that overlaps, to obtain the moving vehicle candidate region of the described relatively first frame vehicle image of the described second frame vehicle image; And
D) in the sorter of sample training gained classified in described moving vehicle candidate region, to detect the moving vehicle zone of the described relatively first frame vehicle image of the described second frame vehicle image, described steps d may further comprise the steps:
D1) size of adjusting described moving vehicle candidate region and described training sample is big or small consistent;
D2) eigenwert of the moving vehicle candidate region of the described adjustment of calculating;
D3) eigenwert and the described training sample characteristic threshold value of more described moving vehicle candidate region are to judge moving vehicle zone described in the described moving vehicle candidate region.
2. vehicle checking method as claimed in claim 1 is characterized in that, also comprises utilizing the vehicle dimension size to filter out the construction zone in the described vehicle candidate region and/or the step in noise zone.
3. vehicle checking method as claimed in claim 1 is characterized in that, also comprises utilizing the vehicle dimension size to determine described vehicle candidate region.
4. vehicle checking method as claimed in claim 2, it is characterized in that, by the pixel region area and the described vehicle candidate region of more described vehicle dimension correspondence, will fall much larger than the construction zone of the pixel region of described vehicle dimension correspondence and/or much smaller than the noise area filter of the pixel region of described vehicle dimension correspondence in the described vehicle candidate region.
5. vehicle checking method as claimed in claim 3, it is characterized in that, pixel region area and described vehicle candidate region by more described vehicle dimension correspondence are defined as described vehicle candidate region with the zone that is substantially equal to the pixel region of described vehicle dimension correspondence in the described vehicle candidate region.
6. vehicle checking method as claimed in claim 1 is characterized in that, described step a may further comprise the steps:
A1) extract predetermined number of pixels zone in the described second frame vehicle image;
A2) obtain corresponding minimum zone with the pixel difference in described predetermined number of pixels zone in the described first frame vehicle image; And
A3) the corresponding minimum area coincidence in the described predetermined number of pixels of translation zone with described pixel difference.
7. vehicle checking method as claimed in claim 1 is characterized in that, described sorter is stacked sorter or tree classifier.
8. a vehicles by overlooking pick-up unit is characterized in that, described device comprises:
Translation unit, the second frame vehicle image that the described translation unit translation first frame vehicle image is adjacent overlaps on world coordinate system;
Statistic unit, described statistic unit utilizes pixel color to add up the vehicle candidate region that obtains respectively in described first frame vehicle image and the described second frame vehicle image, described statistic unit is determined road area and described vehicle candidate region in described first frame vehicle image and the described second frame vehicle image according to the maximum ratio of various pixel color in described first frame vehicle image of adding up and the described second frame vehicle image;
Computing unit, described computing unit carries out image difference calculating to described vehicle candidate region and its corresponding pixel region that overlaps, to obtain the moving vehicle candidate region of the described relatively first frame vehicle image of the described second frame vehicle image; And
Taxon, sorter according to the sample training gained is classified to described moving vehicle candidate region, detecting the moving vehicle zone of the described relatively first frame vehicle image of the described second frame vehicle image, the size that described taxon is adjusted described moving vehicle candidate region is big or small consistent with described training sample; Calculate the eigenwert of the moving vehicle candidate region of described adjustment; And the characteristic threshold value size of the eigenwert of more described moving vehicle candidate region and described training sample, to judge moving vehicle zone described in the described moving vehicle candidate region.
9. vehicle detection apparatus as claimed in claim 8 is characterized in that, also comprises filter element, and described filter element utilizes the vehicle dimension size to filter out construction zone and/or noise zone in the described vehicle candidate region.
10. vehicle detection apparatus as claimed in claim 8 is characterized in that, also comprises determining unit, and described determining unit utilizes the vehicle dimension size to determine described vehicle candidate region.
11. vehicle detection apparatus as claimed in claim 9, it is characterized in that, pixel region area and the described vehicle candidate region of described filter element by more described vehicle dimension correspondence will be fallen much larger than the construction zone of the pixel region of described vehicle dimension correspondence and/or much smaller than the noise area filter of the pixel region of described vehicle dimension correspondence in the described vehicle candidate region.
12. vehicle detection apparatus as claimed in claim 10, it is characterized in that, described determining unit is defined as described vehicle candidate region by the pixel region area and the described vehicle candidate region of more described vehicle dimension correspondence with the zone that is substantially equal to the pixel region of described vehicle dimension correspondence in the described vehicle candidate region.
13. vehicle detection apparatus as claimed in claim 8 is characterized in that, described taxon is stacked sorter or tree classifier.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2009101429367A CN101577052B (en) | 2009-05-14 | 2009-05-14 | Device and method for detecting vehicles by overlooking |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2009101429367A CN101577052B (en) | 2009-05-14 | 2009-05-14 | Device and method for detecting vehicles by overlooking |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101577052A CN101577052A (en) | 2009-11-11 |
| CN101577052B true CN101577052B (en) | 2011-06-08 |
Family
ID=41271984
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2009101429367A Expired - Fee Related CN101577052B (en) | 2009-05-14 | 2009-05-14 | Device and method for detecting vehicles by overlooking |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101577052B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102142090B (en) * | 2011-03-15 | 2013-03-13 | 中国科学技术大学 | Vehicle detection method and system |
| CN103049749B (en) * | 2012-12-30 | 2016-06-29 | 信帧电子技术（北京）有限公司 | The recognition methods again of human body under grid blocks |
| CN109615858A (en) * | 2018-12-21 | 2019-04-12 | 深圳信路通智能技术有限公司 | A kind of intelligent parking behavior judgment method based on deep learning |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2007148907A (en) * | 2005-11-29 | 2007-06-14 | Ikegami Tsushinki Co Ltd | Device, method and program for measuring velocity |
| CN101251927A (en) * | 2008-04-01 | 2008-08-27 | 东南大学 | Vehicle detecting and tracing method based on video technique |
| CN101286239A (en) * | 2008-04-22 | 2008-10-15 | 北京航空航天大学 | Aerial shooting traffic video frequency vehicle rapid checking method |
-
2009
- 2009-05-14 CN CN2009101429367A patent/CN101577052B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2007148907A (en) * | 2005-11-29 | 2007-06-14 | Ikegami Tsushinki Co Ltd | Device, method and program for measuring velocity |
| CN101251927A (en) * | 2008-04-01 | 2008-08-27 | 东南大学 | Vehicle detecting and tracing method based on video technique |
| CN101286239A (en) * | 2008-04-22 | 2008-10-15 | 北京航空航天大学 | Aerial shooting traffic video frequency vehicle rapid checking method |
Non-Patent Citations (1)
| Title |
|---|
| JP特开2007148907A 2007.06.14 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101577052A (en) | 2009-11-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111368687B (en) | Sidewalk vehicle illegal parking detection method based on target detection and semantic segmentation | |
| CN100545867C (en) | Aerial shooting traffic video frequency vehicle rapid checking method | |
| CN111899227A (en) | Automatic railway fastener defect acquisition and identification method based on unmanned aerial vehicle operation | |
| CN103279756B (en) | Vehicle detection based on integrated classifier analyzes system and determination method thereof | |
| CN114170580B (en) | Expressway-oriented abnormal event detection method | |
| CN102855758A (en) | Detection method for vehicle in breach of traffic rules | |
| CN103942560B (en) | A kind of high-resolution video vehicle checking method in intelligent traffic monitoring system | |
| CN109766769A (en) | A kind of road target detection recognition method based on monocular vision and deep learning | |
| CN106128121B (en) | Vehicle queue length fast algorithm of detecting based on Local Features Analysis | |
| CN114049572A (en) | Detection method for identifying small target | |
| CN102902983B (en) | A kind of taxi identification method based on support vector machine | |
| CN104978567A (en) | Vehicle detection method based on scenario classification | |
| CN101714213A (en) | Vehicle-mounted pedestrian detection method and system | |
| CN103679214B (en) | Vehicle checking method based on online Class area estimation and multiple features Decision fusion | |
| CN116434159A (en) | Traffic flow statistics method based on improved YOLO V7 and Deep-Sort | |
| CN102693427A (en) | Method and device for forming detector for detecting images | |
| CN103337175A (en) | Vehicle type recognition system based on real-time video steam | |
| Su et al. | A new local-main-gradient-orientation HOG and contour differences based algorithm for object classification | |
| Yang et al. | Vehicle counting method based on attention mechanism SSD and state detection | |
| CN101577052B (en) | Device and method for detecting vehicles by overlooking | |
| CN116597411A (en) | Method and system for identifying traffic sign by unmanned vehicle in extreme weather | |
| CN110889347A (en) | Density traffic flow counting method and system based on space-time counting characteristics | |
| CN116630904B (en) | Small target vehicle detection method integrating non-adjacent jump connection and multi-scale residual error structure | |
| CN114708560B (en) | YOLOX algorithm-based illegal parking detection method and system | |
| Zhao et al. | An Improved Method for Infrared Vehicle and Pedestrian Detection Based on YOLOv5s |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20110608 Termination date: 20150514 |
|
| EXPY | Termination of patent right or utility model |