CN101477625A - Upper half of human body detection method and system - Google Patents
Upper half of human body detection method and system Download PDFInfo
- Publication number
- CN101477625A CN101477625A CNA2009100762589A CN200910076258A CN101477625A CN 101477625 A CN101477625 A CN 101477625A CN A2009100762589 A CNA2009100762589 A CN A2009100762589A CN 200910076258 A CN200910076258 A CN 200910076258A CN 101477625 A CN101477625 A CN 101477625A
- Authority
- CN
- China
- Prior art keywords
- haar
- candidate window
- little feature
- human body
- upper half
- 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.)
- Granted
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 65
- 238000001914 filtration Methods 0.000 claims abstract description 73
- 238000010606 normalization Methods 0.000 claims abstract description 41
- 238000000034 method Methods 0.000 claims abstract description 28
- 238000012549 training Methods 0.000 claims abstract description 23
- 239000000284 extract Substances 0.000 claims description 54
- 238000012545 processing Methods 0.000 claims description 27
- 238000000605 extraction Methods 0.000 claims description 9
- 230000009466 transformation Effects 0.000 claims description 3
- PICXIOQBANWBIZ-UHFFFAOYSA-N zinc;1-oxidopyridine-2-thione Chemical class [Zn+2].[O-]N1C=CC=CC1=S.[O-]N1C=CC=CC1=S PICXIOQBANWBIZ-UHFFFAOYSA-N 0.000 abstract 1
- 238000012544 monitoring process Methods 0.000 description 14
- 238000010586 diagram Methods 0.000 description 7
- 230000008569 process Effects 0.000 description 6
- 230000000694 effects Effects 0.000 description 4
- 239000012467 final product Substances 0.000 description 4
- 230000008901 benefit Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000035800 maturation Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 238000005728 strengthening Methods 0.000 description 1
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a human upper body detection method and a system. The method comprises the following steps: training positive samples and negative samples of the human upper body to obtain a first classifier, a second classifier and a third classifier, and subjecting all searched candidate windows to the first detection filtering by the first classifier according to Harr micro-characteristics; subjecting the remaining candidate windows after the first detection filtering to gray level normalization, and subjecting all of the candidate windows after the gray level normalization to the second detection filtering by the second classifier according to Harr micro-characteristics; and subjecting all of the remaining candidate windows after the second detection filtering to the third detection filtering by the third classifier according to the distribution of the Harr micro-characteristics so as to filter off non-upper body images not meeting the position shape of the head and shoulders of the human upper body. Accordingly, the method and the system can realize human upper body detection.
Description
Technical field
The present invention relates to Human Detection, particularly a kind of upper half of human body detection method and a kind of upper half of human body detection system.
Background technology
The target that all kinds of video monitorings are paid close attention to mainly comprises two classes, and a class is the automobile as the traffic monitoring target, and another kind of then is people as the indoor and outdoor monitoring objective.
In the prior art, whether certain target that detects and discern in the monitoring scene is the people, features such as profile that can based target, length and width realize, whether monitoring of this class and recognition objective are that people's method is simpler, but the comparatively complicated and more situation of target then is difficult to be suitable for for monitoring scene.
Perhaps, prior art can judge whether certain target in the monitoring scene is the people by the method that only detects the number of people, still, because the information of the number of people is limited, thereby only detects whether certain target that the number of people is difficult to accurately identify in the monitoring scene is the people.
Thus, in order to improve accuracy of detection, prior art can also judge whether certain target in the monitoring scene is the people by the method for human body.Yet, this method needs whole human body all to show in image, for more monitoring scenes of part number such as for example indoor scenes, indoor frequently blocks between men, and can only be shown to usually the people the upper part of the body, be head and shoulder, though thereby this method precision is high is difficult to be suitable for.
As seen, prior art is judging by the method for human body whether certain target in the monitoring scene is man-hour, must require whole human body all to show in image, though thereby precision is high is difficult to be suitable for.
Summary of the invention
In view of this, the invention provides a kind of upper half of human body detection method and a kind of upper half of human body detection system, can realize detection upper half of human body.
A kind of upper half of human body detection method provided by the invention comprises:
Search obtains candidate window in the image of input;
Utilize in advance the first order sorter that obtains according to the positive sample of upper half of human body and anti-sample training, from all candidate window that search obtains, extract little feature of Haar and gray average feature respectively, according to the little feature of Haar and the gray average feature that extract, all candidate window that search obtains are carried out first order detection filtration then;
The first order is detected the remaining candidate window in filtration back carry out the gray scale normalization processing;
Utilize in advance the second level sorter that obtains according to the positive sample of upper half of human body and anti-sample training, extract the little feature of Haar respectively all candidate window after gray scale normalization is handled, according to the little feature of Haar that extracts, all candidate window after the gray scale normalization processing are carried out the second level detect filtration then;
Utilize in advance the third level sorter that obtains according to the positive sample of upper half of human body and anti-sample training, after detecting filtration, the second level extracts the little feature of Haar respectively remaining all candidate window, the regularity of distribution of the little feature of Haar of foundation extraction detects remaining all candidate window in filtration back to the second level and carries out third level detection filtration then;
The third level is detected the upper half of human body that the remaining candidate window in filtration back is defined as comprising the number of people and shoulder.
The little feature of Haar that described first order sorter and described second level sorter extract comprises:
The little feature of first kind Haar, the equal value difference of pixel grey scale between black region that the expression left and right sides is adjacent and the white portion;
The little feature of the second class Haar is represented the equal value difference of pixel grey scale between a neighbouring black region and the white portion;
The little feature of the 3rd class Haar is represented the pixel grey scale average between each adjacent with its left and right sides respectively white portion of black region;
The little feature of the 4th class Haar is represented the equal value difference of pixel grey scale between the black region that two diagonal angles link to each other and the white portion that adjacent two diagonal angles link to each other;
The little feature of the 5th class Haar is represented the equal value difference of pixel grey scale between a black region and the white portion that its diagonal angle, upper right side links to each other;
The equal value difference of pixel grey scale between the little feature of the 6th class Haar, black region and a white portion that its diagonal angle, upper left side links to each other.
The little feature of Haar that described third level sorter extracts comprise first and second, the little feature of five, six class Haar, and the regularity of distribution of the little feature of Haar that extracts of described third level sorter comprises:
First kind Haar little feature absolute value sum in the candidate window in the arbitrary region and in this zone first and second, the merchant of the little feature absolute value of five, six class Haar summation;
Second class Haar little feature absolute value sum in the candidate window in the arbitrary region and in this zone first and second, the merchant of the little feature absolute value of five, six class Haar summation;
The 5th class Haar little feature absolute value sum in the candidate window in the arbitrary region and in this zone first and second, the merchant of the little feature absolute value of five, six class Haar summation;
The 6th class Haar little feature absolute value sum in the candidate window in the arbitrary region and in this zone first and second, the merchant of the little feature absolute value of five, six class Haar summation.
This method sets in advance a upper half of human body masterplate, and described upper half of human body masterplate comprises background area and head shoulder zone;
Described first order sorter and described second level sorter utilize described upper half of human body masterplate, only extract the little feature of first to the 6th class Haar that at least one half-pix point is positioned at described head shoulder zone;
Described third level sorter utilizes described upper half of human body masterplate, only extract at least one half-pix point be positioned at described head shoulder zone first and second, the little feature of five, six class Haar.
Described in the image of input search obtain candidate window and comprise: the image to input carries out the convergent-divergent of preset ratio, the rotation of predetermined angle; The input image and carry out obtaining some candidate window of different size with exhaustive mode search in described convergent-divergent, the described postrotational image; Some candidate window of different size are carried out the size normalized, obtain some candidate window of preset standard size;
Wherein, described preset standard size is consistent with the positive sample of upper half of human body and the anti-sample of upper half of human body of training described first order sorter, second level sorter and third level sorter.
Describedly the third level detect is filtered the remaining candidate window in back be defined as comprising before the upper half of human body of the number of people and shoulder, this method further comprises: the third level is detected filter in remaining all candidate window in back, adjacent a plurality of candidate window are merged into one;
Describedly the third level detect is filtered the remaining candidate window in back be defined as comprising that the upper half of human body of the number of people and shoulder comprises: only the candidate window that described merging is obtained is defined as comprising the upper half of human body of the number of people and shoulder.
Described adjacent a plurality of candidate window comprise: size difference each other is less than presetting second threshold value and/or position difference less than presetting the 3rd threshold value and/or the overlapping area a plurality of candidate window greater than default the 4th threshold value.
A kind of upper half of human body detection system provided by the invention comprises:
The window search unit is used for obtaining candidate window in the image search of input;
The first order sorter that obtains according to the positive sample of upper half of human body and anti-sample training in advance, be used for extracting little feature of Haar and gray average feature respectively from all candidate window that search obtains, and, all candidate window that search obtains are carried out first order detection filtration according to the little feature of Haar and the gray average feature that extract;
The gray scale normalization unit is used for that the first order is detected the remaining candidate window in filtration back and carries out the gray scale normalization processing;
The second level sorter that obtains according to the positive sample of upper half of human body and anti-sample training in advance, all candidate window after being used for handling from gray scale normalization extract the little feature of Haar respectively, according to the little feature of Haar that extracts, all candidate window after the gray scale normalization processing are carried out the second level detect filtration then;
The third level sorter that obtains according to the positive sample of upper half of human body and anti-sample training in advance, be used for detecting remaining all candidate window in filtration back and extract the little feature of Haar respectively from the second level, the regularity of distribution of the little feature of Haar of foundation extraction detects remaining all candidate window in filtration back to the second level and carries out third level detection filtration then;
Identifying unit is used for the third level is detected the upper half of human body that the remaining candidate window in filtration back is defined as comprising the number of people and shoulder as a result.
The little feature of Haar that described first order sorter and described second level sorter extract comprises:
The little feature of first kind Haar, the equal value difference of pixel grey scale between black region that the expression left and right sides is adjacent and the white portion;
The little feature of the second class Haar is represented the equal value difference of pixel grey scale between a neighbouring black region and the white portion;
The little feature of the 3rd class Haar is represented the pixel grey scale average between each adjacent with its left and right sides respectively white portion of black region;
The little feature of the 4th class Haar is represented the equal value difference of pixel grey scale between the black region that two diagonal angles link to each other and the white portion that adjacent two diagonal angles link to each other;
The little feature of the 5th class Haar is represented the equal value difference of pixel grey scale between a black region and the white portion that its diagonal angle, upper right side links to each other;
The equal value difference of pixel grey scale between the little feature of the 6th class Haar, black region and a white portion that its diagonal angle, upper left side links to each other.
The little characteristic distribution rule of the Haar of described third level sorter foundation comprises:
First kind Haar little feature absolute value sum in the candidate window in the arbitrary region and in this zone first and second, the merchant of the little feature absolute value of five, six class Haar summation;
Second class Haar little feature absolute value sum in the candidate window in the arbitrary region and in this zone first and second, the merchant of the little feature absolute value of five, six class Haar summation;
The 5th class Haar little feature absolute value sum in the candidate window in the arbitrary region and in this zone first and second, the merchant of the little feature absolute value of five, six class Haar summation;
The 6th class Haar little feature absolute value sum in the candidate window in the arbitrary region and in this zone first and second, the merchant of the little feature absolute value of five, six class Haar summation.
Further be provided with the upper half of human body masterplate in described first order sorter, second level sorter and the described third level sorter, described upper half of human body masterplate comprises background area and head shoulder zone;
Described first order sorter and described second level sorter utilize described upper half of human body masterplate, only extract the little feature of first to the 6th class Haar that at least one half-pix point is positioned at described head shoulder zone;
Described third level sorter utilizes described upper half of human body masterplate, only extract at least one half-pix point be positioned at described head shoulder zone first and second, the little feature of five, six class Haar.
Described window search unit comprises:
The image transformation subelement is used for the image of input is carried out the convergent-divergent of preset ratio, the rotation of predetermined angle;
The exhaustive search subelement, be used for the input image and carry out described convergent-divergent, described postrotational image, obtain some candidate window of different size with exhaustive mode search;
Size normalization subelement is used for some candidate window of different size are carried out the size normalized, obtains some candidate window of preset standard size;
Wherein, described preset standard size is consistent with the positive sample of upper half of human body and the anti-sample of upper half of human body of training described first order sorter, second level sorter and third level sorter.
This system further comprises the window merge cells at described third level sorter and described as a result between the identifying unit, is used for the third level detected filtering remaining all candidate window in back, and adjacent a plurality of candidate window are merged into one;
And the candidate window that described third level sorter only obtains described merging is defined as comprising the upper half of human body of the number of people and shoulder.
Described adjacent a plurality of candidate window comprise: size difference each other is less than presetting second threshold value and/or position difference less than presetting the 3rd threshold value and/or the overlapping area a plurality of candidate window greater than default the 4th threshold value.
As seen from the above technical solution, the present invention obtains the first order, the second level and third level sorter according to positive sample of upper half of human body and anti-sample training in advance, and all candidate window of utilizing first order sorter by the little feature of Harr search to be obtained are carried out first order detection filtration; After this, the remaining candidate window in first order detection filtration back is carried out gray scale normalization handle, and utilize all candidate window after second level sorter is handled gray scale normalization by the little feature of Harr to carry out second level detection filtration; Then, utilize the regularity of distribution of third level sorter again by the little feature of Haar, remaining all candidate window in filtration back are detected in the second level carry out third level detection filtration, filtering out the non-upper half of human body image that does not satisfy the position shape rule of head and shoulder in the upper half of human body, thereby can realize detection to upper half of human body.
And, because detecting the candidate window of filtration, the first order of carrying out first order sorter do not carry out the gray scale normalization processing, thereby can detect and filter out the comparatively complicated non-upper half of human body image of a large amount of intensity profile, thereby can reduce the processing of second level sorter and third level sorter, and then improve the efficient that upper half of human body detects.
Further, first order sorter, second level sorter and third level sorter all can utilize the upper half of human body masterplate that sets in advance to carry out the little Feature Extraction of Harr, to avoid extracting the little feature of useless Haar in the background area, improve the efficient that upper half of human body detects again further.
In addition, the present invention can also merge adjacent a plurality of candidate window earlier, thereby has avoided the corresponding a plurality of candidate window of same upper half of human body, has further improved the accuracy that upper half of human body detects; And, the appearance of false-alarm is often more isolated owing to the just possible corresponding a plurality of candidate window of real upper half of human body, therefore, if the candidate window that the present invention obtains merging is defined as comprising the upper half of human body of the number of people and shoulder, then can avoid the false-alarm flase drop in the image is surveyed is upper half of human body, thereby has further improved the accuracy that upper half of human body detects.
Description of drawings
Fig. 1 is the composition synoptic diagram of the first order and second level sorter in the embodiment of the invention;
Fig. 2 is the exemplary block diagram of the little feature of Haar in the embodiment of the invention;
Fig. 3 is the exemplary process diagram of upper half of human body detection method in the embodiment of the invention;
Fig. 4 is the synoptic diagram of upper half of human body masterplate in the embodiment of the invention;
Fig. 5 is the exemplary block diagram of upper half of human body detection system in the embodiment of the invention;
Fig. 6 is the structural representation of window search unit in the embodiment of the invention upper half of human body detection system.
Embodiment
For making purpose of the present invention, technical scheme and advantage clearer, below with reference to the accompanying drawing embodiment that develops simultaneously, the present invention is described in more detail.
Since in the more monitoring scene of part numbers such as for example indoor scene, can only be shown to usually the people the upper part of the body, be head and shoulder, therefore, for whether certain target that can effectively judge in this class monitoring scene is the people, present embodiment has proposed the detection method and the system of upper half of human body, like this, a very important person head and shoulder show, can identify the position that this target is behaved and identified this target in image.
Specifically, in the present embodiment, all candidate window that search from input picture is obtained adopt three grades to detect filtration.Wherein, the first order detect to filter utilizes the two class sorters of " upper half of human body/non-upper half of human body " to realize, and at the candidate window of handling without gray scale normalization, carry out the first order by the little feature of Ha Er (Haar) and detect filtration; The second level detect to be filtered and also to be utilized the two class sorters of " upper half of human body/non-upper half of human body " to realize, at the candidate window of handling through gray scale normalization, carry out the second level by the little feature of Haar and detect and filter; And the third level still utilizes upper half of human body/non-upper half of human body " two class sorters realize that the candidate window after detect filtering at the second level but the regularity of distribution by the little feature of Haar rather than the little feature of Haar itself are carried out the third level and detected and filter.
The two class sorters of above-mentioned " upper half of human body/non-upper half of human body " can determine whether the rectangle candidate window of certain yardstick is upper half of human body.If the rectangle candidate window is long is m, wide is n, then correspondingly, the flow process that upper half of human body detects can be in the image of input exhaustive search and to differentiate all sizes be that the window of m * n pixel is as candidate window, each candidate window is input in " upper half of human body/non-upper half of human body " sorter, can stays the candidate window that is identified as upper half of human body.The two class sorters of " upper half of human body/non-upper half of human body " abbreviate " sorter " in this article as.
Required first order sorter, second level sorter, the third level sorter of present embodiment all can utilize the Adaboost theory of maturation in the existing human face detection tech to realize.
Specifically, the AdaBoost theory is the general-purpose algorithm that a kind of Weak Classifier that will be better than conjecture at random arbitrarily is combined into strong classifier, therefore, the present embodiment combination is based on the existing method of the little feature selecting of a kind of Haar of AdaBoost theory, a plurality of Weak Classifiers based on single feature are consisted of a strong classifier, then a plurality of strong classifiers are cascaded into two class sorters of complete " upper half of human body/non-upper half of human body ", i.e. required first order sorter, second level sorter, the third level sorter of present embodiment.
Referring to Fig. 1, first order sorter, second level sorter, third level sorter are formed by the above-mentioned strong classifier cascade of n layer, when first order sorter, second level sorter, the detection of third level sorter, if it is (False) vacation that certain one deck strong classifier in the n layer strong classifier is differentiated a candidate window, then get rid of this window and further do not differentiate, if it is true to be output as (True), the more complicated strong classifier of one deck is differentiated this window under then using.
That is to say that each layer strong classifier can both allow almost all the positive sample of upper half of human body passes through, and refuses the anti-sample of most of non-upper half of human body.The candidate window of input low layer strong classifier is just many like this, and the high-rise candidate window of input significantly reduces.
In addition, for first order sorter, second level sorter, the third level sorter of said structure, also need to utilize positive sample of a large amount of upper half of human body and the anti-sample of upper half of human body to train in advance.
Need to prove that first order sorter can be to train by little feature of Haar that extracts little feature of Haar and gray average feature and extract based on the Adaboost algorithm identified from positive sample of a plurality of upper half of human body and the anti-sample of a plurality of upper half of human body and gray average feature to obtain; Since second level sorter at be the candidate window of handling through gray scale normalization, therefore, the training of second level sorter need not the gray average feature, and can be only to train by the extraction little feature of Haar from positive sample of a plurality of upper half of human body and the anti-sample of a plurality of upper half of human body and based on the little feature of Haar that the Adaboost algorithm identified extracts to obtain; And third level sorter can be to train by the regularity of distribution of the little feature of Haar that extracts the little feature of Haar and extract based on the Adaboost algorithm identified from positive sample of a plurality of upper half of human body and the anti-sample of a plurality of upper half of human body to obtain.
Preferably, in the present embodiment, the little feature of Haar that first order sorter and second level sorter are extracted comprises 6 kinds of little features of Haar shown in Fig. 2 left side, and the gray average feature that first order sorter extracted then is a kind of gray average feature shown in Fig. 2 rightmost side.Above-mentioned 6 kinds of little features of Haar are followed successively by in Fig. 2 from left to right:
The little feature of first kind Haar, the equal value difference of pixel grey scale between black region that the expression left and right sides is adjacent and the white portion (black region is positioned at the right side in Fig. 2, but actual being not limited thereto);
The little feature of the second class Haar is represented the equal value difference of pixel grey scale (black region is positioned at downside in Fig. 2, but actual being not limited thereto) between a neighbouring black region and the white portion;
The little feature of the 3rd class Haar, represent the pixel grey scale average (certainly, also can be between each adjacent with its left and right sides respectively black region of a white portion pixel grey scale average) between each adjacent with its left and right sides respectively white portion of black region;
The little feature of the 4th class Haar is represented the equal value difference of pixel grey scale between the black region that two diagonal angles link to each other and the white portion that adjacent two diagonal angles link to each other;
The little feature of the 5th class Haar is represented the equal value difference of pixel grey scale between a black region and the white portion that its diagonal angle, upper right side links to each other;
The equal value difference of pixel grey scale between the little feature of the 6th class Haar, black region and a white portion that its diagonal angle, upper left side links to each other.
For 6 kinds of little features of Haar as shown in Figure 2, the difference of corresponding black region and white portion interior pixel gray average obtains feature in the present embodiment computed image; For the gray average feature, present embodiment then calculates the average of all pixels in the rectangle frame.
Wherein, the background image of above-mentioned black region ordinary representation upper half of human body, above-mentioned white portion is the ordinary representation upper half of human body then; And in 6 kinds of little features of group as shown in Figure 2, the length and width of black region or white portion can be selected arbitrarily, and the size that only need be no more than candidate window gets final product.
Different with second level sorter with first order sorter is, 6 kinds of little features of Haar that third level sorter in the present embodiment is extracted then can include only the first kind, second class, the 5th class and the little feature of the 6th class Haar in above-mentioned 6 kinds of little features of Haar; And third level sorter has also been considered the little characteristic distribution rule of Haar of any each regional inherent different directions of dividing in the candidate window, and the little characteristic distribution of this Haar can be reacted the boundary intensity of certain regional inherent all directions of upper half of human body.Correspondingly, the little characteristic distribution rule of Haar of third level sorter extraction comprises:
First kind Haar little feature absolute value sum in the candidate window in the arbitrary region and in this zone first and second, the merchant of the little feature absolute value of five, six class Haar summation;
Second class Haar little feature absolute value sum in the candidate window in the arbitrary region and in this zone first and second, the merchant of the little feature absolute value of five, six class Haar summation;
The 5th class Haar little feature absolute value sum in the candidate window in the arbitrary region and in this zone first and second, the merchant of the little feature absolute value of five, six class Haar summation;
The 6th class Haar little feature absolute value sum in the candidate window in the arbitrary region and in this zone first and second, the merchant of the little feature absolute value of five, six class Haar summation.
The above-mentioned regularity of distribution can also adopt following expression mode:
Suppose in the arbitrary region in candidate window that the little feature absolute value of first kind Haar sum, the little feature absolute value of second class Haar sum, the 5th class Haar little feature absolute value sum and the little feature absolute value of the little feature Haar of the 6th class Haar sum are expressed as S
Haar(i), i=0,1,2,3, above-mentioned four the characteristic distribution rules in then should the zone then are expressed as
Wherein, the length and width in selected zone can be selected arbitrarily in the present embodiment, and the size that only need be no more than candidate window gets final product.
And, present embodiment is in equal conditions when processed in order to guarantee the positive sample of all upper half of human body and the anti-sample of upper half of human body, before training, can set the size of sample searches window earlier, for example 19 * 19, sample searches window by first order sorter and second level sorter utilization setting size carries out cutting and size normalized to positive sample of all upper half of human body and the anti-sample of upper half of human body then, obtains positive sample of measure-alike upper half of human body and the anti-sample of upper half of human body.
In addition, present embodiment is for strengthening first order sorter, second level sorter, the different attitude upper half of human body of third level sorter to tilting or rotating to an angle, or the robustness of the upper half of human body of different sizes detection, before the positive sample of all upper half of human body and the anti-sample of upper half of human body are carried out cutting and size normalized, can also carry out mirror image to positive sample of all upper half of human body and the anti-sample of upper half of human body earlier, for example ± 10 degree waits rotation at any angle, for example 1.05 times size is amplified, for example 0.95 times size such as dwindles at processing, to expand the quantity of positive sample of upper half of human body and the anti-sample of upper half of human body.
Below, based on above-mentioned first order sorter, second level sorter and third level sorter, upper half of human body detection method in the present embodiment and upper half of human body detection system are elaborated.
Fig. 3 is the exemplary process diagram of upper half of human body detection method in the embodiment of the invention.As shown in Figure 3, the upper half of human body detection method in the present embodiment comprises:
Preferably, in order to guarantee that as far as possible all possible candidate window can not omitted in the input picture, the processing procedure in this step can specifically comprise:
A1, to the input image carry out the convergent-divergent of preset ratio, the rotation of predetermined angle;
A2, the input image and carry out obtaining some candidate window of different size with exhaustive mode search in described convergent-divergent, the described postrotational image;
A3, some candidate window of different size are carried out the size normalized, obtain some candidate window of preset standard size.
Like this, through the processing of above-mentioned steps a1, a2, can avoid the candidate window of different angles or different sizes to be omitted to greatest extent; Through the processing of above-mentioned a3, also can guarantee in follow-up processing procedure, all candidate window are adopted the processing of equal conditions.
Preferably, when carrying out this step, present embodiment can be enabled an assumed condition, the image that is current input is from video monitoring scene, and the automatic exposure parameter of taking the camera of input picture is a normal numerical value, make entire image bright secretly moderate, brightness is even, this just means, the upper half of human body background image generally is a black in the scene, and people's face in upper half of human body front can be too not black, also can be not bright excessively.Certainly, this step only is can produce better effect for above-mentioned assumed condition, but is not limited in above-mentioned assumed condition.
Because all candidate window are after gray scale normalization is handled, might exist the candidate window of some non-upper half of human body similar in the intensity profile of people face part with actual candidate window for upper half of human body, distinguish comparatively difficulty, therefore, this step does not carry out the gray scale normalization processing and detects the candidate window of filtering some above-mentioned non-upper half of human body by the first order excluding to all candidate window earlier, to reduce follow-up is the processing of distinguishing the candidate window of some above-mentioned non-upper half of human body, thereby can improve the efficient that upper half of human body detects.
Need to prove, in this step from the little feature of Harr that each candidate window extracts the length and width of black region or white portion can select arbitrarily, the size that only need be no more than candidate window gets final product, and for example 1 * 1,3 * 1,1 * 3,1 * 5,5 * 1,2 * 2,2 * 4,4 * 2,3 * 3 etc.; The position of the little feature of Harr also can be selected arbitrarily.
Step 303 detects the remaining candidate window in filtration back to the first order and carries out the gray scale normalization processing.
Step 304 is utilized second level sorter all candidate window after gray scale normalization is handled and is extracted the little feature of Haar respectively, then according to the little feature of Haar that extracts, all candidate window after the gray scale normalization processing is carried out second level detection filter.Wherein, the second level sorter that utilizes of this step can constitute in advance and train and obtains according to foregoing mode; And the little feature of the Haar that extracts in this step can be the little feature of Haar of the foregoing first kind to the six classes.
Preferably, when carrying out this step, also can enable assumed condition mentioned in the step 302.Certainly, this step only is can produce better effect for assumed condition mentioned in the step 302, but is not limited in above-mentioned assumed condition.
Though all candidate window are after gray scale normalization is handled, might exist the candidate window of some non-upper half of human body similar in the intensity profile of people face part, distinguish comparatively difficulty with actual candidate window for upper half of human body, but, the candidate window of some above-mentioned non-upper half of human body is excluded because having detected when filtering for the first time, therefore, the subsequent step that begins from this step has all been avoided the processing to the candidate window of some above-mentioned non-upper half of human body, thereby has improved the efficient that upper half of human body detects.
Need to prove, in this step from the little feature of Harr that each candidate window extracts the length and width of black region or white portion can select arbitrarily, the size that only need be no more than candidate window gets final product, and for example 1 * 1,3 * 1,1 * 3,1 * 5,5 * 1,2 * 2,2 * 4,4 * 2,3 * 3 etc.; The position of the little feature of Harr also can be selected arbitrarily.
Step 305, utilize third level sorter, extract the little feature of Haar respectively from remaining all candidate window in detection filtration back, the second level, the regularity of distribution of the little feature of Haar of foundation extraction is carried out third level detection filtration to remaining all candidate window in detection filtration back, the second level then.Wherein, the third level sorter that utilizes of this step can constitute in advance and train and obtains according to foregoing mode; And the little feature of the Haar that extracts in this step can for foregoing first and second, the little feature of Haar of five, six classes.
Preferably, when carrying out this step, also can enable assumed condition mentioned in the step 302.Certainly, this step only is can produce better effect for assumed condition mentioned in the step 302, but is not limited in above-mentioned assumed condition.
Step 306 is merged into one with adjacent a plurality of candidate window in remaining all candidate window in third level detection filtration back.
Described adjacent can being meant of this step: size difference each other less than pre-set dimension difference threshold value and/or position difference less than predeterminated position difference threshold value and/or overlapping area greater than default overlapping area threshold value.
Because some neighboring candidate window that search obtains from input picture, in fact may be corresponding be same upper half of human body in this input picture, therefore, a plurality of neighboring candidate windows for fear of the same upper half of human body of correspondence are identified as different upper half of human body respectively, by this step adjacent a plurality of candidate window are merged into one and only handled at the candidate window after merging by subsequent step, to improve the accuracy that upper half of human body detects; And, the appearance of false-alarm is often more isolated owing to the just possible corresponding a plurality of candidate window of real upper half of human body, therefore, if subsequent step is only handled at the candidate window after merging, then can avoid the false-alarm flase drop in the image is surveyed is upper half of human body, thereby can improve the accuracy that upper half of human body detects again further.
Certainly, because the effect of this step mainly is to improve the accuracy that upper half of human body detects, do not reduce the accuracy that upper half of human body detects and the realization that can not hinder upper half of human body to detect if do not carry out this step and only be, thereby this step is the step of optional nonessential execution in the practical application, and in Fig. 3, be expressed as frame of broken lines.
Step 307 detects the upper half of human body that the remaining candidate window in filtration back is defined as comprising the number of people and shoulder with the third level.
Need to prove, because step 306 is optional step, therefore, when execution of step 305 back, execution in step 306 and when directly carrying out this step not, this step can directly detect the third level filters the upper half of human body that back remaining all candidate window all be defined as comprising the number of people and shoulder.
After this step, information such as the position of all right all detected upper half of human body of further output, size.
So far, this flow process finishes.
By above-mentioned flow process as seen, upper half of human body detection method in the present embodiment obtains the first order, the second level and third level sorter according to positive sample of upper half of human body and anti-sample training in advance, and all candidate window of utilizing first order sorter by the little feature of Harr search to be obtained are carried out first order detection filtration; After this, the remaining candidate window in first order detection filtration back is carried out gray scale normalization handle, and utilize all candidate window after second level sorter is handled gray scale normalization by the little feature of Harr to carry out second level detection filtration; Then, utilize the regularity of distribution of third level sorter again by the little feature of Haar, remaining all candidate window in filtration back are detected in the second level carry out third level detection filtration, filtering out the non-upper half of human body image that does not satisfy the position shape rule of head and shoulder in the upper half of human body, thereby can realize detection to upper half of human body.
And, because detecting the candidate window of filtration, the first order of carrying out first order sorter do not carry out the gray scale normalization processing, thereby can detect and filter out the comparatively complicated non-upper half of human body image of a large amount of intensity profile, thereby can reduce the processing of second level sorter and third level sorter, and then improve the efficient that upper half of human body detects.
In addition, in order to reduce the treatment capacity of first order sorter, second level sorter and third level sorter, present embodiment is before the above-mentioned flow process of carrying out as shown in Figure 3, one as shown in Figure 4 upper half of human body masterplate can further be set, and this upper half of human body masterplate as shown in Figure 4 comprises background area 401 (black region among Fig. 4) and head shoulder zone 402 (white portions among Fig. 4).
Then when execution in step 302 and step 304, can only extract the little feature of first to the 6th class Haar that at least one half-pix point is positioned at head shoulder zone 402 respectively by first order sorter and second level sorter utilization this upper half of human body masterplate as shown in Figure 4; And when execution in step 305, then then utilize as shown in Figure 4 this upper half of human body masterplate by third level sorter, only extract at least one half-pix point be positioned at head shoulder zone 402 first and second, the little feature of five, six class Haar.Like this, can avoid extracting the little feature of useless Haar in the background area 401, improve the efficient that upper half of human body detects again further.
Fig. 5 is the exemplary block diagram of upper half of human body detection system in the embodiment of the invention.As shown in Figure 5, the upper half of human body detection system in the present embodiment comprises:
Gray scale normalization unit 503 is used for that the first order is detected the remaining candidate window in filtration back and carries out the gray scale normalization processing;
Identifying unit 507 as a result, are used for the third level is detected the upper half of human body that the remaining candidate window in filtration back is defined as comprising the number of people and shoulder.
As shown in Figure 6, in said system, window search unit 501 can specifically comprise:
In addition, still referring to Fig. 5, alternatively, in order to improve the accuracy that upper half of human body detects, said system is at third level sorter 505 and as a result between the identifying unit 507, can further include the window merge cells 506 shown in frame of broken lines among Fig. 5, be used for the third level is detected remaining all candidate window in filtration back, adjacent a plurality of candidate window are merged into one; Wherein, described here adjacent can be meant each other size difference less than pre-set dimension difference threshold value and/or position difference less than predeterminated position difference threshold value and/or overlapping area greater than default overlapping area threshold value.
For the situation that has further comprised window merge cells 506, the candidate window that only merging obtained of identifying unit 507 is defined as upper half of human body as a result.
As seen, upper half of human body detection system in the present embodiment has in advance first order sorter 502, second level sorter 504 and the third level sorter 505 that obtains according to the positive sample of upper half of human body and anti-sample training, and carries out the first order by all candidate window that first order sorter 502 obtains search by the little feature of Harr and detect and filter; After this, the first order detect is filtered the remaining candidate window in back carry out gray scale normalization and handle, and all candidate window after gray scale normalization being handled by the little feature of Harr by second level sorter 504 are carried out the second level and detected and filter; Then, again by the regularity of distribution of third level sorter 505 by the little feature of Haar, remaining all candidate window in filtration back are detected in the second level carry out third level detection filtration, filtering out the non-upper half of human body image that does not satisfy the position shape rule of head and shoulder in the upper half of human body, thereby can realize detection to upper half of human body.
And, because detecting the candidate window of filtration, the first order of carrying out first order sorter 502 do not carry out the gray scale normalization processing, thereby can detect and filter out the comparatively complicated non-face image of a large amount of intensity profile, thereby can reduce the processing of second level sorter 504 and third level sorter 505, and then improve the efficient that upper half of human body detects.
In addition, in order to reduce the treatment capacity of first order sorter 502, second level sorter 504 and third level sorter 505, can further be provided with upper half of human body masterplate as shown in Figure 4 in the first order sorter 502 in the present embodiment upper half of human body detection system, second level sorter 504 and the third level sorter 505.
Like this, this upper half of human body masterplate that first order sorter 502 and second level sorter 504 utilize as shown in Figure 4 only extracts the little feature of first to the 6th class Haar that at least one half-pix point is positioned at head shoulder zone 402; Third level sorter 505 is utilization this upper half of human body masterplate as shown in Figure 4 then, only extracts at least one half-pix point and is positioned at first and second of shoulder zone 402, the little feature of five, six class Haar.
The above is preferred embodiment of the present invention only, is not to be used to limit protection scope of the present invention.Within the spirit and principles in the present invention all, any modification of being done, be equal to and replace and improvement etc., all should be included within protection scope of the present invention.
Claims (14)
1, a kind of upper half of human body detection method is characterized in that, this method comprises:
Search obtains candidate window in the image of input;
Utilize in advance the first order sorter that obtains according to the positive sample of upper half of human body and anti-sample training, from all candidate window that search obtains, extract little feature of Haar and gray average feature respectively, according to the little feature of Haar and the gray average feature that extract, all candidate window that search obtains are carried out first order detection filtration then;
The first order is detected the remaining candidate window in filtration back carry out the gray scale normalization processing;
Utilize in advance the second level sorter that obtains according to the positive sample of upper half of human body and anti-sample training, extract the little feature of Haar respectively all candidate window after gray scale normalization is handled, according to the little feature of Haar that extracts, all candidate window after the gray scale normalization processing are carried out the second level detect filtration then;
Utilize in advance the third level sorter that obtains according to the positive sample of upper half of human body and anti-sample training, after detecting filtration, the second level extracts the little feature of Haar respectively remaining all candidate window, the regularity of distribution of the little feature of Haar of foundation extraction detects remaining all candidate window in filtration back to the second level and carries out third level detection filtration then;
The third level is detected the upper half of human body that the remaining candidate window in filtration back is defined as comprising the number of people and shoulder.
2, the method for claim 1 is characterized in that, the little feature of Haar that described first order sorter and described second level sorter extract comprises:
The little feature of first kind Haar, the equal value difference of pixel grey scale between black region that the expression left and right sides is adjacent and the white portion;
The little feature of the second class Haar is represented the equal value difference of pixel grey scale between a neighbouring black region and the white portion;
The little feature of the 3rd class Haar is represented the pixel grey scale average between each adjacent with its left and right sides respectively white portion of black region;
The little feature of the 4th class Haar is represented the equal value difference of pixel grey scale between the black region that two diagonal angles link to each other and the white portion that adjacent two diagonal angles link to each other;
The little feature of the 5th class Haar is represented the equal value difference of pixel grey scale between a black region and the white portion that its diagonal angle, upper right side links to each other;
The equal value difference of pixel grey scale between the little feature of the 6th class Haar, black region and a white portion that its diagonal angle, upper left side links to each other.
3, method as claimed in claim 2 is characterized in that, the little feature of Haar that described third level sorter extracts comprise first and second, the little feature of five, six class Haar, and the regularity of distribution of the little feature of Haar that extracts of described third level sorter comprises:
First kind Haar little feature absolute value sum in the candidate window in the arbitrary region and in this zone first and second, the merchant of the little feature absolute value of five, six class Haar summation;
Second class Haar little feature absolute value sum in the candidate window in the arbitrary region and in this zone first and second, the merchant of the little feature absolute value of five, six class Haar summation;
The 5th class Haar little feature absolute value sum in the candidate window in the arbitrary region and in this zone first and second, the merchant of the little feature absolute value of five, six class Haar summation;
The 6th class Haar little feature absolute value sum in the candidate window in the arbitrary region and in this zone first and second, the merchant of the little feature absolute value of five, six class Haar summation.
4, method as claimed in claim 3 is characterized in that, this method sets in advance a upper half of human body masterplate, and described upper half of human body masterplate comprises background area and head shoulder zone;
Described first order sorter and described second level sorter utilize described upper half of human body masterplate, only extract the little feature of first to the 6th class Haar that at least one half-pix point is positioned at described head shoulder zone;
Described third level sorter utilizes described upper half of human body masterplate, only extract at least one half-pix point be positioned at described head shoulder zone first and second, the little feature of five, six class Haar.
5, as each described method in the claim 1 to 4, it is characterized in that,
Described in the image of input search obtain candidate window and comprise: the image to input carries out the convergent-divergent of preset ratio, the rotation of predetermined angle; The input image and carry out obtaining some candidate window of different size with exhaustive mode search in described convergent-divergent, the described postrotational image; Some candidate window of different size are carried out the size normalized, obtain some candidate window of preset standard size;
Wherein, described preset standard size is consistent with the positive sample of upper half of human body and the anti-sample of upper half of human body of training described first order sorter, second level sorter and third level sorter.
6, method as claimed in claim 5, it is characterized in that, describedly the third level detect is filtered the remaining candidate window in back be defined as comprising before the upper half of human body of the number of people and shoulder, this method further comprises: the third level is detected filter in remaining all candidate window in back, adjacent a plurality of candidate window are merged into one;
Describedly the third level detect is filtered the remaining candidate window in back be defined as comprising that the upper half of human body of the number of people and shoulder comprises: only the candidate window that described merging is obtained is defined as comprising the upper half of human body of the number of people and shoulder.
7, method as claimed in claim 6, it is characterized in that described adjacent a plurality of candidate window comprise: size difference each other is less than presetting second threshold value and/or position difference less than presetting the 3rd threshold value and/or the overlapping area a plurality of candidate window greater than default the 4th threshold value.
8, a kind of upper half of human body detection system is characterized in that, comprising:
The window search unit is used for obtaining candidate window in the image search of input;
The first order sorter that obtains according to the positive sample of upper half of human body and anti-sample training in advance, be used for extracting little feature of Haar and gray average feature respectively from all candidate window that search obtains, and, all candidate window that search obtains are carried out first order detection filtration according to the little feature of Haar and the gray average feature that extract;
The gray scale normalization unit is used for that the first order is detected the remaining candidate window in filtration back and carries out the gray scale normalization processing;
The second level sorter that obtains according to the positive sample of upper half of human body and anti-sample training in advance, all candidate window after being used for handling from gray scale normalization extract the little feature of Haar respectively, according to the little feature of Haar that extracts, all candidate window after the gray scale normalization processing are carried out the second level detect filtration then;
The third level sorter that obtains according to the positive sample of upper half of human body and anti-sample training in advance, be used for detecting remaining all candidate window in filtration back and extract the little feature of Haar respectively from the second level, the regularity of distribution of the little feature of Haar of foundation extraction detects remaining all candidate window in filtration back to the second level and carries out third level detection filtration then;
Identifying unit is used for the third level is detected the upper half of human body that the remaining candidate window in filtration back is defined as comprising the number of people and shoulder as a result.
9, system as claimed in claim 8 is characterized in that, the little feature of Haar that described first order sorter and described second level sorter extract comprises:
The little feature of first kind Haar, the equal value difference of pixel grey scale between black region that the expression left and right sides is adjacent and the white portion;
The little feature of the second class Haar is represented the equal value difference of pixel grey scale between a neighbouring black region and the white portion;
The little feature of the 3rd class Haar is represented the pixel grey scale average between each adjacent with its left and right sides respectively white portion of black region;
The little feature of the 4th class Haar is represented the equal value difference of pixel grey scale between the black region that two diagonal angles link to each other and the white portion that adjacent two diagonal angles link to each other;
The little feature of the 5th class Haar is represented the equal value difference of pixel grey scale between a black region and the white portion that its diagonal angle, upper right side links to each other;
The equal value difference of pixel grey scale between the little feature of the 6th class Haar, black region and a white portion that its diagonal angle, upper left side links to each other.
10, system as claimed in claim 9 is characterized in that, the little characteristic distribution rule of the Haar of described third level sorter foundation comprises:
First kind Haar little feature absolute value sum in the candidate window in the arbitrary region and in this zone first and second, the merchant of the little feature absolute value of five, six class Haar summation;
Second class Haar little feature absolute value sum in the candidate window in the arbitrary region and in this zone first and second, the merchant of the little feature absolute value of five, six class Haar summation;
The 5th class Haar little feature absolute value sum in the candidate window in the arbitrary region and in this zone first and second, the merchant of the little feature absolute value of five, six class Haar summation;
The 6th class Haar little feature absolute value sum in the candidate window in the arbitrary region and in this zone first and second, the merchant of the little feature absolute value of five, six class Haar summation.
11, system as claimed in claim 10, it is characterized in that, further be provided with the upper half of human body masterplate in described first order sorter, second level sorter and the described third level sorter, described upper half of human body masterplate comprises background area and head shoulder zone;
Described first order sorter and described second level sorter utilize described upper half of human body masterplate, only extract the little feature of first to the 6th class Haar that at least one half-pix point is positioned at described head shoulder zone;
Described third level sorter utilizes described upper half of human body masterplate, only extract at least one half-pix point be positioned at described head shoulder zone first and second, the little feature of five, six class Haar.
As each described system in the claim 8 to 11, it is characterized in that 12, described window search unit comprises:
The image transformation subelement is used for the image of input is carried out the convergent-divergent of preset ratio, the rotation of predetermined angle;
The exhaustive search subelement, be used for the input image and carry out described convergent-divergent, described postrotational image, obtain some candidate window of different size with exhaustive mode search;
Size normalization subelement is used for some candidate window of different size are carried out the size normalized, obtains some candidate window of preset standard size;
Wherein, described preset standard size is consistent with the positive sample of upper half of human body and the anti-sample of upper half of human body of training described first order sorter, second level sorter and third level sorter.
13, system as claimed in claim 12, it is characterized in that, this system is at described third level sorter and described as a result between the identifying unit, further comprise the window merge cells, be used for the third level is detected remaining all candidate window in filtration back, adjacent a plurality of candidate window are merged into one;
And the candidate window that described third level sorter only obtains described merging is defined as comprising the upper half of human body of the number of people and shoulder.
14, system as claimed in claim 13, it is characterized in that described adjacent a plurality of candidate window comprise: size difference each other is less than presetting second threshold value and/or position difference less than presetting the 3rd threshold value and/or the overlapping area a plurality of candidate window greater than default the 4th threshold value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200910076258 CN101477625B (en) | 2009-01-07 | 2009-01-07 | Upper half of human body detection method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 200910076258 CN101477625B (en) | 2009-01-07 | 2009-01-07 | Upper half of human body detection method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101477625A true CN101477625A (en) | 2009-07-08 |
| CN101477625B CN101477625B (en) | 2013-04-24 |
Family
ID=40838334
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 200910076258 Active CN101477625B (en) | 2009-01-07 | 2009-01-07 | Upper half of human body detection method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101477625B (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101630410B (en) * | 2009-08-18 | 2011-08-17 | 北京航空航天大学 | Human body sitting posture judgment method based on single camera |
| CN102184388A (en) * | 2011-05-16 | 2011-09-14 | 苏州两江科技有限公司 | Face and vehicle adaptive rapid detection system and detection method |
| CN102324041A (en) * | 2011-09-09 | 2012-01-18 | 深圳泰山在线科技有限公司 | Pixel classification method, joint body gesture recognition method and mouse instruction generating method |
| US9020188B2 (en) | 2011-05-23 | 2015-04-28 | Asustek Computer Inc. | Method for object detection and apparatus using the same |
| CN105303163A (en) * | 2015-09-22 | 2016-02-03 | 深圳市华尊科技股份有限公司 | Method and detection device for target detection |
| CN105488468A (en) * | 2015-11-26 | 2016-04-13 | 浙江宇视科技有限公司 | Method and device for positioning target area |
| CN106599781A (en) * | 2016-11-08 | 2017-04-26 | 国网山东省电力公司威海供电公司 | Electric power business hall dressing normalization identification method based on color and Hu moment matching |
| CN106815596A (en) * | 2016-12-08 | 2017-06-09 | 中国银联股份有限公司 | A kind of Image Classifier method for building up and device |
| CN106886216A (en) * | 2017-01-16 | 2017-06-23 | 深圳前海勇艺达机器人有限公司 | Robot automatic tracking method and system based on RGBD Face datections |
| CN107028427A (en) * | 2017-06-08 | 2017-08-11 | 翼猫科技发展（上海）有限公司 | The method for sensing of the sleep system of new automated intelligent regulation |
| CN107122791A (en) * | 2017-03-15 | 2017-09-01 | 国网山东省电力公司威海供电公司 | Electricity business hall employee's hair style specification detection method based on color development and Texture Matching |
| CN107633232A (en) * | 2017-09-26 | 2018-01-26 | 四川长虹电器股份有限公司 | A kind of low-dimensional faceform's training method based on deep learning |
| CN110807494A (en) * | 2019-11-08 | 2020-02-18 | 深圳市深视创新科技有限公司 | Rapid positioning method for repeated textures in industrial vision |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN100565556C (en) * | 2004-05-14 | 2009-12-02 | 欧姆龙株式会社 | Specific object detection device |
| CN100405388C (en) * | 2004-05-14 | 2008-07-23 | 欧姆龙株式会社 | Detector for special shooted objects |
| CN101320484B (en) * | 2008-07-17 | 2012-01-04 | 清华大学 | Three-dimensional human face recognition method based on human face full-automatic positioning |
-
2009
- 2009-01-07 CN CN 200910076258 patent/CN101477625B/en active Active
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101630410B (en) * | 2009-08-18 | 2011-08-17 | 北京航空航天大学 | Human body sitting posture judgment method based on single camera |
| CN102184388A (en) * | 2011-05-16 | 2011-09-14 | 苏州两江科技有限公司 | Face and vehicle adaptive rapid detection system and detection method |
| US9020188B2 (en) | 2011-05-23 | 2015-04-28 | Asustek Computer Inc. | Method for object detection and apparatus using the same |
| CN102324041A (en) * | 2011-09-09 | 2012-01-18 | 深圳泰山在线科技有限公司 | Pixel classification method, joint body gesture recognition method and mouse instruction generating method |
| CN102324041B (en) * | 2011-09-09 | 2014-12-03 | 深圳泰山在线科技有限公司 | Pixel classification method, joint body gesture recognition method and mouse instruction generating method |
| CN105303163B (en) * | 2015-09-22 | 2019-03-01 | 深圳市华尊科技股份有限公司 | A kind of method and detection device of target detection |
| CN105303163A (en) * | 2015-09-22 | 2016-02-03 | 深圳市华尊科技股份有限公司 | Method and detection device for target detection |
| CN105488468A (en) * | 2015-11-26 | 2016-04-13 | 浙江宇视科技有限公司 | Method and device for positioning target area |
| CN105488468B (en) * | 2015-11-26 | 2019-10-18 | 浙江宇视科技有限公司 | A kind of localization method and device of target area |
| CN106599781A (en) * | 2016-11-08 | 2017-04-26 | 国网山东省电力公司威海供电公司 | Electric power business hall dressing normalization identification method based on color and Hu moment matching |
| CN106815596A (en) * | 2016-12-08 | 2017-06-09 | 中国银联股份有限公司 | A kind of Image Classifier method for building up and device |
| CN106886216A (en) * | 2017-01-16 | 2017-06-23 | 深圳前海勇艺达机器人有限公司 | Robot automatic tracking method and system based on RGBD Face datections |
| CN107122791A (en) * | 2017-03-15 | 2017-09-01 | 国网山东省电力公司威海供电公司 | Electricity business hall employee's hair style specification detection method based on color development and Texture Matching |
| CN107028427A (en) * | 2017-06-08 | 2017-08-11 | 翼猫科技发展（上海）有限公司 | The method for sensing of the sleep system of new automated intelligent regulation |
| CN107028427B (en) * | 2017-06-08 | 2020-02-21 | 翼猫科技发展（上海）有限公司 | Novel sensing method of sleep system capable of automatically and intelligently adjusting |
| CN107633232A (en) * | 2017-09-26 | 2018-01-26 | 四川长虹电器股份有限公司 | A kind of low-dimensional faceform's training method based on deep learning |
| CN110807494A (en) * | 2019-11-08 | 2020-02-18 | 深圳市深视创新科技有限公司 | Rapid positioning method for repeated textures in industrial vision |
| CN110807494B (en) * | 2019-11-08 | 2023-05-30 | 深圳市深视创新科技有限公司 | Quick positioning method for repeated textures in industrial vision |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101477625B (en) | 2013-04-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101477625B (en) | Upper half of human body detection method and system | |
| CN101447023B (en) | Method and system for detecting human head | |
| CN106682601B (en) | A kind of driver's violation call detection method based on multidimensional information Fusion Features | |
| CN101271514B (en) | Image detection method and device for fast object detection and objective output | |
| CN101777114B (en) | Intelligent analysis system and intelligent analysis method for video monitoring, and system and method for detecting and tracking head and shoulder | |
| US9390333B2 (en) | System and method for robust motion detection | |
| US7492926B2 (en) | Method for identifying a person from a detected eye image | |
| CN111191573A (en) | Driver fatigue detection method based on blink rule recognition | |
| CN106295551A (en) | A kind of personal security cap wear condition real-time detection method based on video analysis | |
| CN101477641A (en) | Demographic method and system based on video monitoring | |
| EP1675066A1 (en) | Face image candidate area search method, face image candidate area search system, and face image candidate area search program | |
| JP5106356B2 (en) | Image monitoring device | |
| CN106339657B (en) | Crop straw burning monitoring method based on monitor video, device | |
| CN101546375B (en) | Face detection method with characteristic reduction | |
| CN105095837B (en) | A kind of TV station symbol recognition method and system | |
| CN105740752B (en) | Sensitive picture filtering method and system | |
| KR100755800B1 (en) | Face detector and detecting method using facial color and adaboost | |
| CN102521610B (en) | Image filtering method and device | |
| Arafah et al. | Face recognition system using Viola Jones, histograms of oriented gradients and multi-class support vector machine | |
| US8958598B2 (en) | System and method for detecting moving objects using video capturing device | |
| CN107341456B (en) | Weather sunny and cloudy classification method based on single outdoor color image | |
| CN107403192B (en) | Multi-classifier-based rapid target detection method and system | |
| CN115294632A (en) | True and false smile identification method and system | |
| CN114882401A (en) | Flame detection method and system based on RGB-HSI model and flame initial growth characteristics | |
| Lausser et al. | Detecting zebra crossings utilizing AdaBoost. |
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 | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20210203 Address after: No. 602, 6th floor, shining building, 35 Xueyuan Road, Haidian District, Beijing 100083 Patentee after: BEIJING ZHONGXINGTIANSHI TECHNOLOGY Co.,Ltd. Address before: 100083, Haidian District, Xueyuan Road, Beijing No. 35, Nanjing Ning building, 15 Floor Patentee before: Vimicro Corp. |
|
| TR01 | Transfer of patent right |