CN101604380B - Method for identifying human head by diameter searching - Google Patents
Method for identifying human head by diameter searching Download PDFInfo
- Publication number
- CN101604380B CN101604380B CN200910055122XA CN200910055122A CN101604380B CN 101604380 B CN101604380 B CN 101604380B CN 200910055122X A CN200910055122X A CN 200910055122XA CN 200910055122 A CN200910055122 A CN 200910055122A CN 101604380 B CN101604380 B CN 101604380B
- Authority
- CN
- China
- Prior art keywords
- people
- pixel point
- white pixel
- search
- image
- 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
The invention relates to a method for identifying a human head by diameter searching. Video collection equipment is arranged on a position at certain distance from top of the head of a person, and shoots from top to bottom; the collecting human head is basically a circle-similar area; and each pixel point of an image is searched by a diameter searching method; the circle-similar characteristic of the area is determined from several directions so as to identify the human head and obtain the center and the radius of the human head. The method avoids complicated mathematical formulae, and abstract spatial switching concept in Hough transformation, improves accuracy, and has the characteristics of small occupied memory and high real time.
Description
Technical field
The present invention relates to a kind of image processing techniques, particularly a kind of method for identifying human head by diameter searching.
Background technology
The detection of mankind's activity has been played key effect in many-sides such as automatic video monitoring, man-machine interaction are used.In utilizing computer implemented automatic recognition system, people's head detected be considered to detect people's activity one of the most effective means, because the information of head is to be discerned by computer system the easiliest.Robustness that is used to detect the automatic recognition system of mankind's activity directly depends on accuracy and the real-time to head part's identification.Nowadays, head detection has been applied in many fields, for example at pedestrian's detection with at the detection of ridership.To the detection of ridership, camera is installed in number of people top usually, takes from top to down.The main task that the number of people detects is to determine the size and location of number of people part.Realtime graphic by camera collection is used as input, and automatic recognition system is handled it, and output is about mathematical description information such as the number that whether has head or head in the image, positions.
Camera is taken from top to down, and the number of people image of acquisition is exactly a similar round target.People's head inspecting method based on this environment has developed a lot, and they each have its limitation.For example, utilize in conjunction with gradient and color histogram and carry out number of people identification, this method recognition accuracy is subjected to having a strong impact on of clothes color.Four marginal points are randomly drawed in employing, judge whether to exist a possible circular target by the distance of calculating them, and this method is under the bigger environment of noise, and hit rate is very low, thereby influence its detection speed.The fuzzy C-means clustering method is used owing to determining that at first clusters number is difficult under the situation of this clusters number random variation.The most frequently used method of detection type circular target is exactly Hough conversion (HT).HT is owing to have very strong robustness to insensitive for noise and to discontinuous edge, and is used widely.But the Hough conversion not only takies a large amount of internal memories, and to having the detection of three-dimensional parameter space circular target, the real-time problem of HT seems very outstanding.Developed a lot of raising HT speed afterwards and reduced the algorithm of memory demand, as random Hough transformation (RHT), based on the Hough conversion (GHT) of gradient information etc.These improved Hough conversion all are difficult to solve the real-time problem when noise is bigger.
Summary of the invention
The present invention be directed to the circumscribed problem of present number of people identification detection method, a kind of method for identifying human head by diameter searching has been proposed, overcoming the low real-time of classic method, low noise immunity and the big shortcoming of EMS memory occupation amount, is people's head inspecting method of a kind of high-accuracy, high real-time.
Technical scheme of the present invention is: a kind of method for identifying human head by diameter searching, video capture device is installed on a distance, the crown of people, and take from top to bottom, the laggard pedestrian's head identification of image data, method for identifying human head by diameter searching comprises following concrete steps:
1), initialization array A[M] each element value of [N] is 0, A[M] [N] be used for the development length of vertical search white pixel point of each pixel of document image, M presentation video horizontal pixel number wherein, N represents vertical pixel count;
2), the image of the binaryzation of lining by line scan, when meeting white pixel point (x, y) time, program begins vertical detection (x, y+1), (x, y+2) ... in order there to be the cavity in the admissible region, a permission is set at the number T that does not vertically occur white pixel point continuously, the value of T is determined according to the cavity size of allowing existence, when the number that white pixel point vertically continuously do not occur during less than T, continue toward vertically searching for the white pixel point, up to vertically the number of white pixel point not occurring greater than T, at this moment continuously, will (x y) locates to deposit array element A[x in toward the length of longitudinal extension] [y]; After the scanning entire image, just obtained array A[M like this] each element value of [N], it has write down the length of each white pixel point to longitudinal extension, and black pixel point longitudinal extension length has been initialized as 0;
3), search A[M] maximal value of [N], be made as A[x0] [y0], set the minimum value D of number of people diameter in image according to actual conditions, if A[x0] [y0]＞D, then may be at (x0, y0) locate to have a number of people toward longitudinal extension, its home position is (x0, y0+A[x0] [y0]/2), for further checking this place whether to have the number of people, the place is laterally detected after the same method at home position, if circle centre position white pixel point horizontal expansion length also greater than D, then it is that the possibility of the number of people becomes big; Substantially can think that there is the number of people in this place, number of people center is (x0, y0+A[x0] [y0]/2), and radius is A[x0] [y0]/2; Confirm further if desired whether this place is the number of people, can also detect white pixel point development length toward 45 degree directions and 135 degree directions, see whether it meets the requirement of number of people similar round feature at circle centre position;
4), before the next number of people of search, earlier with A[x0] [y0] and array element zero clearing on every side thereof, with get rid of search this time may repeat search to first number of people, search for A[M again] maximal value of [N], be made as A[x1] [y1], (x1 y1) locates whether to exist the number of people according to the method validation of step 3);
5), repeat the 4th) step, the maximal value of coming out up to search shows that less than D the number of people in the image is all found out.
Beneficial effect of the present invention is: the number of people of the present invention identification diameter search procedure, avoid space conversion notion abstract in complex mathematical formula and the Hough conversion, and improved accuracy rate, have that EMS memory occupation is little, the characteristics of high real-time.
Description of drawings
Fig. 1 is the hardware block diagram of number of people identification diameter search procedure of the present invention program run;
Fig. 2 asks for A[M in the number of people of the present invention identification diameter search procedure] process flow diagram of each element of [N] array;
Fig. 3 is according to A[M in the number of people of the present invention identification diameter search procedure] [N] ask for the process flow diagram of number of people parameter.
Embodiment
Camera is taken the situation of carrying out number of people identification from top to down and is applied in usually in the passenger flow statistical system, as supermarket, park, bus etc.Usually adopt the method for background subtraction to extract human body target in this case, adopt quick fuzzy C mean algorithm to ask for gray threshold to the image behind the background subtraction then, carry out binaryzation, carry out number of people identification and counting again.White pixel point after the binaryzation is people head's mark and some noises.
Algorithm of the present invention according to binaryzation after the feature of image design, avoid space conversion notion abstract in complex mathematical formula and the Hough conversion.Because the gray-scale value of the number of people is more concentrated, people's head region that the image after background subtracts is left is more concentrated, and according to the difference of threshold value, there is cavitation in people's head region, will carry out morphology and handle in a lot of algorithms.This algorithm is handled morphology and is merged, and cardinal principle is the diameter search.If the binary image upper left corner is true origin, laterally be directions X, vertically be the Y direction.Each pixel of searching image at first line by line, (establishing coordinate is (x when running into a white pixel point, y)), the direction of search transfers to vertically, and (x y+1) is the white pixel point in detection, owing to there is cavitation, what time can allow continuous is not the white pixel point, searches for toward vertical always, and the quantity that up to continuity point is not white pixel point is above certain threshold value.Note from (x y) begins the down length of search.So just obtain each white pixel point of view picture binary image at development length longitudinally, search for the peaked position of development length then, be made as (x0, y0), this point might be number of people position, development length according to this point finds possible center then, carries out Horizon Search white pixel point again in the center, laterally whether satisfies the similar round feature of the number of people to judge this zone.For further judging whether this zone satisfies number of people feature, can also search for the white pixel point along 45 degree directions and 135 degree directions in the center, carrying out the similar round signature verification of the number of people, is strip target or other non-similar round target to get rid of detected target.Search also verify after first number of people, searches for second possible number of people again, is lower than a certain threshold value (this threshold value is set to the number of people diameter value of minimum) up to the development length of white pixel point.
Hardware block diagram as shown in Figure 1, the hardware configuration of algorithm operation of the present invention comprises: the DM642 hardware circuit board of video acquisition module, TI company and result output module.Since computer vision handle to as if the big digital picture of data volume, so select the high performance DSP that is exclusively used in digital media applications for use, promptly select for use the TMS320DM642 of TI company to deal with device.Video capture device is installed on a distance, the crown of people, takes from top to bottom, the number of people of collection is class circle zone basically.Judge the class circle characteristic in zone from several directions, thereby discern the number of people, and obtain the center and the radius of the number of people.
Binaryzation is asked for A[M as shown in Figure 2] process flow diagram of each element of [N] array and according to A[M] [N] ask for the process flow diagram of number of people parameter, it is as follows to embody concrete algorithm steps:
1), initialization array A[M] each element value of [N] is 0, A[M] [N] be used for the development length of vertical search white pixel point of each pixel of document image, M presentation video horizontal pixel number wherein, N represents vertical pixel count.
2), the line by line scan image of binaryzation, when meeting white pixel point (x, y) time, program begin vertical detection (x, y+1), (x, y+2) ... in order there to be the cavity in the admissible region, a permission is set at the number T that does not vertically occur white pixel point continuously, when vertically the continuous number that white pixel point do not occur is less than T, continue toward vertically searching for the white pixel point, up to vertically the number of white pixel point not occurring greater than T continuously.At this moment, incite somebody to action (x y) locates to deposit array element A[x in toward the length of longitudinal extension] [y].The value of T determines that according to the cavity size of allowing existence its essence of setting T value is handled morphology exactly and has been fused to this algorithm.It is appropriate that the value of T will be selected, and T does not pass by in the too little cavity of people's head region that will cause more; T is too big, will cause searching number of people zone in addition.After the traversal entire image, just obtained array A[M like this] each element value of [N], it has write down the length (black pixel point longitudinal extension length be initialized as 0) of each white pixel point to longitudinal extension.
3), search A[M] maximal value of [N], be made as A[x0] [y0], set the minimum value D of number of people diameter in image according to actual conditions, if A[x0] [y0]＞D, then may be in that (x0 y0) locates to have a number of people toward longitudinal extension, its home position is (x0, y0+A[x0] [y0]/2).For further checking this place whether to have the number of people, the place is laterally detected after the same method at home position, if circle centre position white pixel point horizontal expansion length is also greater than D, then it is that the possibility of the number of people becomes big, substantially can think that there is the number of people in this place, number of people center is (x0, y0+A[x0] [y0]/2), and radius is A[x0] [y0]/2.Confirm further if desired whether this place is the number of people, can also detect white pixel point development length toward 45 degree directions and 135 degree directions, see whether it meets the requirement of number of people similar round feature at circle centre position.
4), before the next number of people of search, earlier with A[x0] [y0] and array element zero clearing on every side thereof, with get rid of search this time may repeat search to first number of people.Search for A[M again] maximal value of [N], be made as A[x1] [y1], (x1 y1) locates whether to exist the number of people according to the method validation of step 3).
5), repeat the 4th) step, the maximal value of coming out up to search shows that less than D the number of people in the image is all found out.
In this number of people detection algorithm, determine A[M] process flow diagram of [N] value, most crucial part is to determine array A[M in the algorithm] [N] each element value, and also be code section consuming time relatively, its core code is as follows:
for(i＝0；i＜M；i++)
For (j=0; J＜N; J++) // travel through whole binary image
{
A[i] [j]=0; // this array writes down each pixel and vertically searches for development length,
If (grey scale pixel value==255 of the capable j of i row) // run into white pixel point
{
Temp=0; The number of white pixel point does not vertically appear in // initialization continuously
For (k=1; K＜N; K++) // vertically search
{
If (i+k＜N﹠amp; ﹠amp; The grey scale pixel value of the capable j row of i+k)==255)
Temp=0; // if white pixel point point then resets
else {
Temp++; // if not the white pixel point, then temp adds one
if(temp＞T)break；
// if marginal point do not appear continuously and outnumber threshold value T, then stop vertically search
}
}
A[i] [j]=k; // note the length of this white pixel point place longitudinal extension
}
}
After having determined the longitudinal extension length A [M] [N] of each pixel, need search A[M] people's header in [N], its process flow diagram is as shown in Figure 3.Program is moved on TMS320DM642, detects the number of people, M=720, N=576 with the diameter search procedure; T=4, the core code execution time is 0.01145 second; Use the Hough change detection, core code working time is 0.49251 second, fast about 40 times of travelling speed.Use different algorithms, the number of people center and the radius that detect also have certain error, but error is very little, has only the error of several pixels.
Claims (1)
1. method for identifying human head by diameter searching is installed on a distance, the crown of people with video capture device, takes from top to bottom, and the laggard pedestrian's head identification of image data, method for identifying human head by diameter searching comprises following concrete steps:
1), initialization array A[M] each element value of [N] is 0, A[M] [N] be used for the development length of vertical search white pixel point of each pixel of document image, M presentation video horizontal pixel number wherein, N represents vertical pixel count;
2), the image of the binaryzation of lining by line scan, when meeting white pixel point (x, y) time, program begins vertical detection (x, y+1), (x, y+2) ... in order there to be the cavity in the admissible region, a permission is set at the number T that does not vertically occur white pixel point continuously, the value of T is determined according to the cavity size of allowing existence, when the number that white pixel point vertically continuously do not occur during less than T, continue toward vertically searching for the white pixel point, up to vertically the number of white pixel point not occurring greater than T, at this moment continuously, will (x y) locates to deposit array element A[x in toward the length of longitudinal extension] [y]; After the scanning entire image, just obtained array A[M like this] each element value of [N], it has write down the length of each white pixel point to longitudinal extension, and black pixel point longitudinal extension length has been initialized as 0;
3), search A[M] maximal value of [N], be made as A[x0] [y0], set the minimum value D of number of people diameter in image according to actual conditions, work as A[x0] [y0]＞D, then at (x0, y0) locating to have a number of people, its home position toward longitudinal extension is (x0, y0+A[x0] [y0]/2), for further checking this place whether to have the number of people, the place is laterally detected after the same method at home position, when circle centre position white pixel point horizontal expansion length also greater than D, then it is that the possibility of the number of people becomes big; Think that there is the number of people in this place, number of people center is (x0, y0+A[x0] [y0]/2), and radius is A[x0] [y0]/2; When needs confirm further whether this place is the number of people, then detect white pixel point development length toward 45 degree directions and 135 degree directions at circle centre position, see whether it meets the requirement of number of people similar round feature;
4), before the next number of people of search, earlier with A[x0] [y0] and array element zero clearing on every side thereof, with get rid of search this time may repeat search to first number of people, search for A[M again] maximal value of [N], be made as A[x1] [y1], (x1 y1) locates whether to exist the number of people according to the method validation of step 3);
5), repeat the 4th) step, the maximal value of coming out up to search shows that less than D the number of people in the image is all found out.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN200910055122XA CN101604380B (en) | 2009-07-21 | 2009-07-21 | Method for identifying human head by diameter searching |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN200910055122XA CN101604380B (en) | 2009-07-21 | 2009-07-21 | Method for identifying human head by diameter searching |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101604380A CN101604380A (en) | 2009-12-16 |
| CN101604380B true CN101604380B (en) | 2011-07-20 |
Family
ID=41470103
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN200910055122XA Expired - Fee Related CN101604380B (en) | 2009-07-21 | 2009-07-21 | Method for identifying human head by diameter searching |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101604380B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2012066910A1 (en) | 2010-11-19 | 2012-05-24 | 株式会社ニコン | Guidance system, detection device, and position assessment device |
| CN103903284A (en) * | 2014-03-24 | 2014-07-02 | 中山新诺科技股份有限公司 | Multi-circle detecting method based on least square method |
| CN105404852B (en) * | 2015-10-28 | 2019-01-25 | 广州视源电子科技股份有限公司 | A kind of method and device showing public restroom vacancy |
| CN106022226B (en) * | 2016-05-11 | 2019-03-01 | 同济大学 | A kind of pedestrian based on multi-direction multichannel strip structure discrimination method again |
| CN107742355B (en) * | 2017-11-13 | 2023-08-18 | 江苏海洋大学 | Automatic device of conveying letter sorting coin |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1334439A (en) * | 2000-06-01 | 2002-02-06 | 株式会社三丰 | Spot image related optical position sensor |
| CN101144708A (en) * | 2007-09-26 | 2008-03-19 | 东南大学 | Three-dimensional scanning system circular index point detection method |
| CN101334263A (en) * | 2008-07-22 | 2008-12-31 | 东南大学 | Circular target circular center positioning method |
| EP2079054A1 (en) * | 2008-01-11 | 2009-07-15 | OMG Plc. | Detection of blobs in images |
-
2009
- 2009-07-21 CN CN200910055122XA patent/CN101604380B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1334439A (en) * | 2000-06-01 | 2002-02-06 | 株式会社三丰 | Spot image related optical position sensor |
| CN101144708A (en) * | 2007-09-26 | 2008-03-19 | 东南大学 | Three-dimensional scanning system circular index point detection method |
| EP2079054A1 (en) * | 2008-01-11 | 2009-07-15 | OMG Plc. | Detection of blobs in images |
| CN101334263A (en) * | 2008-07-22 | 2008-12-31 | 东南大学 | Circular target circular center positioning method |
Non-Patent Citations (1)
| Title |
|---|
| JP特开2007-172642A 2007.07.05 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101604380A (en) | 2009-12-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110427905B (en) | Pedestrian tracking method, device and terminal | |
| CN106951879B (en) | Multi-feature fusion vehicle detection method based on camera and millimeter wave radar | |
| CN109633676A (en) | A kind of method and system based on the laser radar obstruction detection direction of motion | |
| KR100201739B1 (en) | Method for observing an object, apparatus for observing an object using said method, apparatus for measuring traffic flow and apparatus for observing a parking lot | |
| CN102479327B (en) | Target identification system, and the obstacle identification system and method for automobile | |
| JP6226368B2 (en) | Vehicle monitoring apparatus and vehicle monitoring method | |
| CN101604380B (en) | Method for identifying human head by diameter searching | |
| CN109871732B (en) | Parking grid identification system and method thereof | |
| CN104303193A (en) | Clustering-based object classification | |
| Guerreiro et al. | Connectivity-enforcing Hough transform for the robust extraction of line segments | |
| CN110298323B (en) | Frame-fighting detection method, system and device based on video analysis | |
| CN102855461A (en) | Method and equipment for detecting fingers in images | |
| CN114782499A (en) | Image static area extraction method and device based on optical flow and view geometric constraint | |
| Wu et al. | Traffic monitoring and vehicle tracking using roadside cameras | |
| Zhang et al. | Bidirectional multiple object tracking based on trajectory criteria in satellite videos | |
| CN110096934A (en) | Road surface identification detection method and equipment | |
| CN109101874B (en) | Library robot obstacle identification method based on depth image | |
| CN111830966A (en) | Corner recognition and cleaning method, device and storage medium | |
| US8116524B2 (en) | Method of object detection | |
| Michael et al. | Fast change detection for camera-based surveillance systems | |
| CN116110230A (en) | Vehicle lane crossing line identification method and system based on vehicle-mounted camera | |
| CN105930813B (en) | A method of detection composes a piece of writing this under any natural scene | |
| KR100976142B1 (en) | detection method of road vehicles | |
| JP4842301B2 (en) | Pedestrian detection device and program | |
| Wang et al. | A system of automated training sample generation for visual-based car detection |
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: 20110720 Termination date: 20140721 |
|
| EXPY | Termination of patent right or utility model |