CN101645137A - Method for automatically detecting location of a football in long shot of football video - Google Patents
Method for automatically detecting location of a football in long shot of football video Download PDFInfo
- Publication number
- CN101645137A CN101645137A CN 200910089438 CN200910089438A CN101645137A CN 101645137 A CN101645137 A CN 101645137A CN 200910089438 CN200910089438 CN 200910089438 CN 200910089438 A CN200910089438 A CN 200910089438A CN 101645137 A CN101645137 A CN 101645137A
- Authority
- CN
- China
- Prior art keywords
- football
- area
- frame
- long shot
- video
- 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
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention provides a method for automatically detecting the location of a football in the long shot of a football video. The method comprises the following steps: firstly, decoding the long shot part in a football video, to acquire successive long-shot frames; processing the images of the long-shot frames, so as to extract the football field part; then, creating the binary saliency map of theprocessed images by integrating color information; further detecting the binary saliency map by the detection features, such as the shape and dimension of the football, so as to locate all the objectssimilar to the football; and analyzing and predicting the objects similar to the football in the current frame by the football location in the previous frame, so as to locate the football. By integrating various spatial information and using the temporal correlation of the successive frames at the same time, the method of the invention has higher accuracy of detection results; furthermore, the entire detection process is carried out on a real-time basis, thereby fully meeting the requirements for the application in the real-time live broadcast of football videos on a portable terminal and ensuring higher application values.
Description
Technical field
The present invention relates in football video automatic detection to the football position, specifically, relate to in the long shot of football video to the automatic testing method of football position.
Background technology
Football has vast participation always and pay close attention to colony, and the huge commercial value that it contains also makes people that football video is being carried out constantly research as one of most popular ball sports items in the whole world.Wherein the automatic analysis to football video is a main aspect, specifically comprises aspects such as retrieval to video, summary, detection, reinforcement, and then carries out the understanding on the semantic hierarchies, such as tactics with go up form analysis or the like.And, just become one of above-mentioned analysis at first to need the problem that solves for the detection of football position in view of the focus of football on always, and the application after also will directly affecting the accuracy of its detection.On the other hand, popular along with portable mobile terminal can watch football match live whenever and wherever possible on portable terminal, and vast football fan is had very strong temptation undoubtedly.But on the small screen, appreciate football match, the long shot in the match particularly, spectators' visual experience and view and admire and experience and to ensure.A method that solves is spectators' area-of-interest in the displaying video picture, and this area-of-interest also to be position with football be center.
Football video has the characteristics of himself as a kind of concrete video classification.Activity such as sportsman and ball is all carried out on the football field basically, and there is specific color in the court.Such as football concrete big or small color shape is arranged also.In addition, the classification of football video camera lens generally has three kinds: long shot, medium shot and close shot camera lens.Wherein, in medium shot and close shot camera lens, football all is bigger, and the position is also relatively easily determined.And in long shot, the size of football is very little with respect to whole camera lens, in the very fast camera lens of some motions, what football-shaped neither sub-circular.And with respect to football, the background object in the camera lens is also very complicated, has introduced a lot of noises and can cause strong interference to the detection of football.A lot of detection methods in the past can't meet the demands in this case.Yet in our application, in the football video of the small screen is appreciated, mainly be exactly that long shot causes difficulty partly for spectators' understanding, the method for solution also is only to play the area-of-interest part of long shot picture.
In addition, current football detection technique all is to detect at relatively large football, just how can the quite good detecting result in medium shot.And for long shot, the littler shape of football size is more irregular, and background is complicated more, and noise is more serious, and the power that is detected as of these methods just is difficult to guarantee.And there have the circular hough transform of the special utilization of a class to detect circular football detection method to be just more inapplicable in long shot.In addition, in the broadcast of the portable terminal live interest that can attract the user with football match, and one of live requirement is a real-time at football video.Also having the detection method of a class football is the track that first easy detection some frames are followed the tracks of football again, and this method can't satisfy live requirement on real-time.
Summary of the invention
The object of the present invention is to provide a kind ofly, detect in the long shot of football video of football position automatic testing method automatically the football position at the long shot in the football video.This detection method integrating various spatial information is utilized successive frame correlativity in time, guarantees that testing result has higher accuracy.And whole testing process all is to carry out in real time, meets football video real-time live application requirements on portable terminal device fully, has bigger using value.
In order to realize above purpose, to the automatic testing method of football position, it is characterized in that in the long shot of a kind of football video of the present invention, comprise the steps:
1) long shot in the football video is partly decoded, obtain continuous long shot frame;
Usually, present football video has all passed through mpeg, has H.264 waited coding, utilizes ffmpeg just can be decoded into continuous frame.As for to the determining of lens type, golden section is carried out in the zone of present frame, in each zone investigation its belong to the shared ratio of meadow color pixel.
2) the long shot two field picture is handled, is extracted the court part:
According to (for example to continuous a plurality of frames, 30 frames) statistics of the main areas color on the picture obtains main color, because the major part of big multiframe is exactly the court in the football video, just can determines the cardinal principle color in court according to this main color value, and then extract the court part of image;
Here, for example can adopt the hsv color model,, also be suitable for differentiating different colors because the hsv color model follows people's perception to suit more.Its color component (Hue) can independently show the color under the different brightness situations, so just can get rid of the interference of court in different weather, light situation.
3) set up handling the binary saliency map (Binary Saliency Map) of image afterwards:
Based on people's visual system feature, will concentrate into measuring of a unification to the feature description of the several different aspects of color;
Here, can utilize brightness, red green contrast characteristic and champac contrast characteristic to make up remarkable figure.In addition, the direction on the picture, texture information feature etc. also can suitably be used.
The respectively following formula of above-mentioned brightness, red green contrast characteristic and champac contrast characteristic's extracting method calculates:
Wherein Θ be picture (d represents the difference of two eigenwerts for i, the j) neighborhood of position (in long shot, we generally take the neighborhood of 3x3 size), I, R, G, B and Y represent brightness, red, green, blue, the yellow several characteristic of this location of pixels respectively.These three characteristic weighings are comprehensive:
S(i,j)=λ
IS
1(i,j)+λ
RGS
RG(i,j)+λ
BYS
BY(i,j)
Wherein, λ
1, λ
RG, λ
BYIt is weighting factor.
In general, brightness and yellowish green contrast characteristic are important, because the color of ball is approximate yellow or white substantially, and the color in court is green partially.Yet when emphasizing these two features, certain part of the sportsman in court is divided the possibility that comes can be bigger, can cause interference to the detection of ball.Therefore, three weighting factors are made as equivalence here.
Thus, obtain the remarkable figure of this two field picture, just obtained binary saliency map after filling through denoising, gray scale adjustment, binaryzation and zone afterwards.
4) on this binary saliency map, detect according to the detected characteristics of ball, find out the approximate object of all footballs;
Here, the detected characteristics of ball specifically comprises:
A) size: at first eliminating can not be the obvious excessive of ball or cross undersized object, determine the possible scope of football size according to the mean size of residue object then, generally choose 1/2nd to two times big minizone of mean value, adopt the method for statistical learning afterwards, the testing result of preceding 20 frames of every continuous 100 frames is used for renewal and obtains ball size more accurately;
B) position: all generally all are detection noise near the position on border too, and the upper left upper right fixed position of picture generally is the icon of time and rebroadcast television, and these positions are got rid of;
C) ratio of inspected object region area and minimum boundary rectangle:
R
Region_to_MBR=Area
Region/Area
MBR
Wherein, Area
RegionRepresent the area in inspected object zone, Area
MBRRepresent the area of minimum boundary rectangle,
This ratio R
Region_to_MBRUsually more than or equal to 0.6, as be lower than this value and then think it to be football, got rid of;
D) inspected object region area and minimum external polygonal ratio: be used to investigate the compact of the shape of examined object, with circular similarity,
R
Region_to_MBP=Area
Region/Area
MBP
Wherein, Area
RegionRepresent the area in inspected object zone, Area
MBPRepresent minimum external polygonal area,
This ratio R
Region_to_MBPUsually more than or equal to 0.8, as be lower than this value and then think it to be football, got rid of;
E) object area and the circular degree of approximation: be used for inspected object at similar football whether in shape,
R
C_to_MBR=Area
C/Area
MBR
Wherein,
Here, the Area_circle representative is the border circular areas in the center of circle with the barycenter of object area, and its area and object area area equate, obtain its radius:
Area
CThe object that detected of representative is the area within the homalographic border circular areas Area_circle that its barycenter overlaps, Area
MBRRepresent the area of minimum extraneous rectangle,
This object area and circular degree of approximation R
C_to_MBRUsually more than or equal to 0.55, as be lower than this value and then think it to be football, got rid of;
Above-mentioned c), d) and e) three characteristic indexs be shape facility,
According to above each feature, the object on the court is checked eliminating one by one, left all objects are as the approximate object of football.
5) based on the testing result of step 4), and the testing result of frame before utilizing, handle at the approximate object of all footballs that detect, to determining or predict in the position of ball:
If a) detect more than the approximate thing of a football, then these objects are compared respectively in location and size with the detected football that comes of frame before respectively, obtain position and difference sum in shape: D=D
1+ D
2, D wherein
1Be meant the approximate thing that detects and the position distance of former frame, D
2Be meant area discrepancy between the two, find out the approximate thing of the minimum object of difference sum D value then, then enter situation b) as the best;
B) if detect the approximate thing of a football, then the detected football that comes of this object and frame is before further judged in location and size, if the distance of position on the longitudinal axis and X direction of this object and the detected football of frame before is respectively less than 1/10th of the height of this two field picture and width; And the dimensioned area of this object judges that then this object is football between 1/3rd to three times of the area of the detected ball that comes, otherwise, if do not satisfy as above all conditions, then enter situation c);
C) if being the neither one object, the result who detects is similar to football, then according to before the testing result of all frames, dope the football position of present frame by Kalman filtering, on this predicted position or just be defined as the position range place of football with the close larger object of sportsman with regard to the sportsman of near position or regular shape and size; If the predicted position annex does not have object to exist, then with previous frame football position frame position for this reason, back one frame restarts to detect; If near the body form the predicted position is irregular and size differs bigger, show that then football is blocked by other objects, judge that predicted position is the football position, back one frame restarts to detect.
Because on the court, a lot of situation balls all are to overlap or covered by the sportsman with the sportsman, and it is very difficult finding accurate football position like this, also there is no need simultaneously.It is pointed out that must to have near the predicted position object existence could can determine the football position range at last, otherwise just specifies former frame to detect to such an extent that the football position is the football position of present frame, and next frame restarts new detection again.
Beneficial effect to the automatic testing method of football position in the long shot of football video of the present invention is: earlier the long shot frame is handled, extract its court part, comprehensive then colouring information, set up binary saliency map, again in information such as the shape that detects object on this binary saliency map, sizes, find out the approximate thing of ball, utilization is analyzed the approximate thing of the football of present frame and is predicted in the detected football of frame position before more afterwards, finds the position of football.Detection method of the present invention combines various spatial informations, has also utilized successive frame correlativity in time, makes testing result have higher accuracy.And whole testing process is all carried out in real time, also meets football video real-time live application requirements on portable terminal device fully, has bigger using value.
Description of drawings
Fig. 1 is the overview flow chart of the automatic testing method of football video long shot mesopodium ball position of the present invention.
Fig. 2 is the particular flow sheet of the football position being judged or being predicted according to the image detection result in the automatic testing method of football video long shot mesopodium ball position of the present invention.
Embodiment
Describe in further detail below in conjunction with accompanying drawing and concrete embodiment automatic testing method football video long shot mesopodium ball position of the present invention.
Fig. 1 is the automatic testing method overview flow chart of football video long shot mesopodium ball position of the present invention.Fig. 2 is the particular flow sheet of the football position being judged or being predicted according to the image detection result in the automatic testing method of football video long shot mesopodium ball position of the present invention.
As depicted in figs. 1 and 2, the automatic testing method of football video long shot mesopodium ball position of the present invention at the long shot frame, at first carries out pre-service, extracts the court part, sets up binary saliency map.Then the object on the binary map is detected one by one, find out the more alike approximate thing of talipes calcaneus ball on the size shape, then these testing results are carried out analyzing and processing, the position of decision football.Specifically comprise the steps:
1) long shot in the football video is partly decoded, obtain continuous long shot frame;
Usually, present football video has all passed through mpeg, has H.264 waited coding, utilizes ffmpeg just can be decoded into continuous frame.As for to the determining of lens type, golden section is carried out in the zone of present frame, in each zone investigation its belong to the shared ratio of meadow color pixel.
2) the long shot two field picture is handled, is extracted the court part:
For example, obtain main color,, just can determine the cardinal principle color in court, and then extract the court part of image according to this main color value because the major part of big multiframe is exactly the court in the football video according to statistics to the main areas color on the continuous 30 frame pictures; Here, extract the method for court part, can adopt the hsv color model,, also be suitable for differentiating different colors because the hsv color model follows people's perception to suit more.Its color component (Hue) can independently show the color under the different brightness situations, so just can get rid of the interference of court in different weather, light situation.For example, can adopt document Keewon Seo and Jaeseung Ko; " An intelligent display schemeof soccer video on mobile devices; " in IEEE Transactions On Circuits and Systems ForVideo Technology, vol.17, NO.10, the method for describing among the Oct.2007: in the hsv color model, add up the property of the histogram of court master's color, obtain the HSV decision method of court color.And then be unit with every 16x16 block of pixels, do judgment basis with certain threshold value, each block of pixels is divided into court part piece and non-court piece.According to the situation around the block of pixels, whether decision is positioned at the court at last.So just, the court part of picture and non-court part can have simply been determined.
3) set up handling the binary saliency map (Binary Saliency Map) of image afterwards:
Based on people's visual system feature, will concentrate into measuring of a unification to the feature description of the several different aspects of color;
Here, can utilize brightness, red green contrast characteristic and champac contrast characteristic to make up remarkable figure.In addition, the direction on the picture, texture information feature etc. also can suitably be used.
In the present embodiment, adopt brightness, red green contrast characteristic and champac contrast characteristic to make up remarkable figure.At this moment, the respectively following formula of above-mentioned brightness, red green contrast characteristic and champac contrast characteristic's extracting method calculates:
Wherein Θ be picture (d represents the difference of two eigenwerts for i, the j) neighborhood of position (in long shot, we generally take the neighborhood of 3x3 size), I, R, G, B and Y represent brightness, red, green, blue, the yellow several characteristic of this location of pixels respectively.These three characteristic weighings are comprehensive:
S(i,j)=λ
IS
1(i,j)+λ
RGS
RG(i,j)+λ
BYS
BY(i,j)
Wherein, λ
I, λ
RG, λ
BYIt is weighting factor.
In general, brightness and yellowish green contrast characteristic are important, because the color of ball is approximate yellow or white substantially, and the color in court is green partially.Yet when emphasizing these two features, certain part of the sportsman in court is divided the possibility that comes can be bigger, can cause interference to the detection of ball.Therefore, three weighting factors are made as equivalence here.
Thus, obtain the remarkable figure of this two field picture, just obtained binary saliency map after filling through denoising, gray scale adjustment, binaryzation and zone afterwards.
4) on described binary saliency map, detect according to the detected characteristics of ball, find out the approximate object of all footballs, finding on each frame may be the object of ball;
Here, the detected characteristics of ball comprises:
A) size: at first eliminating can not be the obvious excessive of ball or cross undersized object, determine the possible scope of football size according to the mean size of residue object then, generally choose 1/2nd to two times big minizone of mean value, adopt the method for statistical learning afterwards, the testing result of preceding 20 frames of every continuous 100 frames is used for renewal and obtains ball size more accurately;
Here, at first utilize dimension information, size obviously can not be the object (oversize or too small) of ball on this frame picture of eliminating earlier, and then try to achieve the mean value of remaining object size, utilize this mean value to determine the possible interval of ball size (being size greater than half of this mean value), ball is carried out preliminary detection get rid of less than a times of mean value.
B) position: all generally all are detection noise near the position on border too, and the upper left upper right fixed position of picture generally is the icon of time and rebroadcast television, and these positions are got rid of;
C) ratio of inspected object region area and minimum boundary rectangle:
R
Region_to_MBR=Area
Region/Area
MBR
Wherein, Area
RegionRepresent the area in inspected object zone, Area
MBRRepresent the area of minimum boundary rectangle,
This ratio R
Region_to_MBRUsually more than or equal to 0.6, as be lower than this value and then think it to be football, got rid of;
D) inspected object region area and minimum external polygonal ratio: be used to investigate the compact of the shape of examined object, with circular similarity,
R
Region_to_MBP=Area
Region/Area
MBP
Wherein, Area
RegionRepresent the area in inspected object zone, Area
MBPRepresent minimum external polygonal area,
This ratio R
Region_to_MBPUsually more than or equal to 0.8, as be lower than this value and then think it to be football, got rid of;
E) object area and the circular degree of approximation: be used for inspected object at similar football whether in shape,
R
C_to_MBR=Area
C/Area
MBR
Wherein,
Here, the Area_circle representative is the border circular areas in the center of circle with the barycenter of object area, and its area and object area area equate, obtain its radius:
Area
CThe object that detected of representative is the area within the homalographic border circular areas Area_circle that its barycenter overlaps, Area
MBRRepresent the area of minimum extraneous rectangle,
This object area and circular degree of approximation R
C_to_MBRUsually more than or equal to 0.55, as be lower than this value and then think it to be football, got rid of;
Above-mentioned c), d) and e) three characteristic indexs be shape facility,
According to above each feature, the object on the court is checked eliminating one by one, left all objects are as the approximate object of football.
Above-mentioned steps 4) in, detection method to the court mainly is to rely on size, position and shape facility etc., the present invention is at foundation a) size and b) on the basis of position feature, further utilize shape facility: c) ratio of inspected object region area and minimum boundary rectangle, d) inspected object region area and minimum external polygonal ratio and e) object area and the circular degree of approximation, object on the court is checked eliminating one by one, determine it to be the object of ball.The per 100 frame detection meetings of meeting of the present invention are added up testing result again, draw nearest ball size data, rely on these data ball to be detected more accurately again.
5) based on the testing result of step 4), and the testing result of frame before utilizing, handle at the approximate object of all footballs that detect, to determining or predict in the position of ball:
If a) detect more than the approximate thing of a football, then these objects are compared respectively in location and size with the detected football that comes of frame before respectively, obtain position and difference sum in shape: D=D
1+ D
2, D wherein
1Be meant the approximate thing that detects and the position distance of former frame, D
2Be meant area discrepancy between the two, find out the approximate thing of the minimum object of difference sum D value then, then enter situation b) as the best;
B) if detect the approximate thing of a football, then the detected football that comes of this object and frame is before further judged in location and size, if the distance of position on the longitudinal axis and X direction of this object and the detected football of frame before is respectively less than 1/10th of the height of this two field picture and width; And the dimensioned area of this object judges that then this object is football between 1/3rd to three times of the area of the detected ball that comes, otherwise, if do not satisfy as above all conditions, then enter situation c);
C) if being the neither one object, the result who detects is similar to football, then according to before the testing result of all frames, dope the football position of present frame by Kalman filtering, on this predicted position or just be defined as the position range place of football with the close larger object of sportsman with regard to the sportsman of near position or regular shape and size; If the predicted position annex does not have object to exist, then with previous frame football position frame position for this reason, back one frame restarts to detect; If near the body form the predicted position is irregular and size differs bigger, show that then football is blocked by other objects, judge that predicted position is the football position, back one frame restarts to detect.
Because on the court, a lot of situation balls all are to overlap or covered by the sportsman with the sportsman, and it is very difficult finding accurate football position like this, also there is no need simultaneously.It is pointed out that must to have near the predicted position object existence could can determine the football position range at last, otherwise just specifies former frame to detect to such an extent that the football position is the football position of present frame, and next frame restarts new detection again.
Fig. 2 is the particular flow sheet of the football position being judged or being predicted according to the image detection result in the automatic testing method of football video long shot mesopodium ball position of the present invention.As shown in Figure 2, this algorithm structure block diagram has been described in the step 5) judgement or the Forecasting Methodology to the football position more accurately.
Wherein, because the motion of football can be regarded a typical Markov process as on the court, present frame football position is the approximate thing of all footballs can be compared with the detected football position of former frame, draws current best object space.This judgment condition is exactly the difference of position and shape: D=D
1+ D
2
D
1Be meant the object that detects and the position distance of former frame, D
2Be meant area discrepancy between the two.
As shown in Figure 2, can't detect the picture of football for court situation complexity, football gross distortion or other--promptly when in the equation D is excessive as a result the time, before at first relying on frame detect football information, Kalman filtering dopes the football position of present frame.On this position or with regard to sportsman or other sizes regular object close with the sportsman of near position all may be the scope place of football, and the present invention is considered as ball and sportsman's integral position the position range of ball.If but size is very big, and may be other detection noise, as detected position, next frame restarts to detect again predicted position.Must must there be the object existence can determine scope at last near it is pointed out that predicted position.Otherwise just directly adopt the testing result of former frame, restart this testing process afterwards again.The appointment of doing these detection positions is can't detect successful camera lens at minority, and specified as can be seen position all near the football position, that is to say it is in the area-of-interest of long shot, so also is in order to be more convenient for this application.
Claims (5)
1, in a kind of long shot of football video to the automatic testing method of football position, it is characterized in that, comprise the steps:
1) long shot in the football video is partly decoded, obtain continuous long shot frame;
2) the long shot two field picture is handled, is extracted the court part:
Obtain main color according to statistics, determine the cardinal principle color in court, and then extract the court part of image according to this main color value to the main areas color on continuous a plurality of frame pictures;
3) set up handling the binary saliency map of image afterwards:
Based on people's visual system feature, will concentrate into measuring of a unification to the feature description of the several different aspects of color;
4) on this binary saliency map, detect according to the detected characteristics of ball, find out the approximate object of all footballs;
5) based on the testing result of step 4), and the testing result of frame before utilizing, handle at the approximate object of all footballs that detect, to determining or predict in the position of ball:
If a) detect more than the approximate thing of a football, then these objects are compared respectively in location and size with the detected football that comes of frame before respectively, obtain position and difference sum in shape: D=D
1+ D
2, D wherein
1Be meant the approximate thing that detects and the position distance of former frame, D
2Be meant area discrepancy between the two, find out the approximate thing of the minimum object of difference sum D value then, then enter situation b) as the best;
B) if detect the approximate thing of a football, then the detected football that comes of this object and frame is before further judged in location and size, if the distance of position on the longitudinal axis and X direction of this object and the detected football of frame before is respectively less than 1/10th of the height of this two field picture and width; And the dimensioned area of this object judges that then this object is football between 1/3rd to three times of the area of the detected ball that comes, otherwise, if do not satisfy as above all conditions, then enter situation c);
C) if being the neither one object, the result who detects is similar to football, then according to before the testing result of all frames, dope the football position of present frame by Kalman filtering, on this predicted position or just be defined as the position range place of football with the close larger object of sportsman with regard to the sportsman of near position or regular shape and size; If the predicted position annex does not have object to exist, then with previous frame football position frame position for this reason, back one frame restarts to detect; If near the body form the predicted position is irregular and size differs bigger, show that then football is blocked by other objects, judge that predicted position is the football position, back one frame restarts to detect.
2, in the long shot of football video as claimed in claim 1 to the automatic testing method of football position, it is characterized in that described step 2) in, adopt the hsv color model.
3, in the long shot of football video as claimed in claim 1 to the automatic testing method of football position, it is characterized in that in the described step 3), utilize brightness, red green contrast characteristic and champac contrast characteristic to make up remarkable figure, concrete grammar is as follows:
Above-mentioned brightness, red green contrast characteristic and champac contrast characteristic's extracting method is following formula respectively:
Wherein Θ is the neighborhood of picture position, and d represents the difference of two eigenwerts, I, and R, G, B and Y represent brightness, red, green, blue, the yellow several characteristic of this location of pixels respectively,
Then, above-mentioned three characteristic weighings are comprehensive:
S(i,j)=λ
IS
I(i,j)+λ
RGS
RG(i,j)+λ
BYS
BY(i,j)
Wherein, λ
I, λ
RG, λ
BYBe weighting factor,
Thus, obtain the remarkable figure of this two field picture, just obtain binary saliency map after filling through denoising, gray scale adjustment, binaryzation and zone afterwards.
4, in the long shot of football video as claimed in claim 3 to the automatic testing method of football position, it is characterized in that described three weighting factor λ
I, λ
RG, λ
BYBe made as equivalence.
5, in the long shot of football video as claimed in claim 1 to the automatic testing method of football position, it is characterized in that in the described step 4), described detected characteristics comprises:
A) size: at first eliminating can not be the obvious excessive of ball or cross undersized object, determine the possible scope of football size according to the mean size of residue object then, generally choose 1/2nd to two times big minizone of mean value, adopt the method for statistical learning afterwards, the testing result of preceding 20 frames of every continuous 100 frames is used for renewal and obtains ball size more accurately;
B) position: all generally all are detection noise near the position on border too, and the upper left upper right fixed position of picture generally is the icon of time and rebroadcast television, and these positions are got rid of;
C) ratio of inspected object region area and minimum boundary rectangle;
R
Re?gion_to_MBR=Area
Re?gion/Area
MBR
Wherein, Area
Re gionRepresent the area in inspected object zone, Area
MBRRepresent the area of minimum boundary rectangle,
This ratio R
Re gion_to_MBRUsually more than or equal to 0.6, as be lower than this value and then think it to be football, got rid of;
D) inspected object region area and minimum external polygonal ratio: be used to investigate the compact of the shape of examined object, with circular similarity,
R
Re?gion_to_MBP=Area
Re?gion/Area
MBP
Wherein, Area
Re gionRepresent the area in inspected object zone, Area
MBPRepresent minimum external polygonal area,
This ratio R
Re gion_to_MBPUsually more than or equal to 0.8, as be lower than this value and then think it to be football, got rid of;
E) object area and the circular degree of approximation: be used for inspected object at similar football whether in shape,
R
C_to_MBR=Area
C/Area
MBR
Wherein,
Here, the Area_circle representative is the border circular areas in the center of circle with the barycenter of object area, and its area and object area area equate, obtain its radius:
Area
CThe object that detected of representative is the area within the homalographic border circular areas Area_circle that its barycenter overlaps, Area
MBRRepresent the area of its minimum extraneous rectangle,
This object area and circular degree of approximation R
C_to_MBRUsually more than or equal to 0.55, as be lower than this value and then think it to be football, got rid of;
Above-mentioned c), d) and e) three characteristic indexs be shape facility,
According to above each feature, the object on the court is checked eliminating one by one, left all objects are as the approximate object of football.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN 200910089438 CN101645137B (en) | 2009-07-17 | 2009-07-17 | Method for automatically detecting location of a football in long shot of football video |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN 200910089438 CN101645137B (en) | 2009-07-17 | 2009-07-17 | Method for automatically detecting location of a football in long shot of football video |
Publications (2)
Publication Number | Publication Date |
---|---|
CN101645137A true CN101645137A (en) | 2010-02-10 |
CN101645137B CN101645137B (en) | 2012-07-04 |
Family
ID=41657018
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN 200910089438 Expired - Fee Related CN101645137B (en) | 2009-07-17 | 2009-07-17 | Method for automatically detecting location of a football in long shot of football video |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN101645137B (en) |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101794451A (en) * | 2010-03-12 | 2010-08-04 | 上海交通大学 | Tracing method based on motion track |
CN102005053A (en) * | 2010-11-17 | 2011-04-06 | 中国科学院声学研究所 | Player position automatic detection method |
CN104778459A (en) * | 2015-04-17 | 2015-07-15 | 广西科技大学 | Feature fusion method for actions of multiple athletes in football match video |
CN104794488A (en) * | 2015-04-17 | 2015-07-22 | 广西科技大学 | Transfer learning method of team performance classification indicated by sparse prototype |
CN104866853A (en) * | 2015-04-17 | 2015-08-26 | 广西科技大学 | Method for extracting behavior characteristics of multiple athletes in football match video |
CN106651952A (en) * | 2016-10-27 | 2017-05-10 | 深圳锐取信息技术股份有限公司 | Football detecting and tracking based video processing method and device |
CN106991358A (en) * | 2016-01-20 | 2017-07-28 | 上海慧体网络科技有限公司 | The algorithm that automatic identification football match based on panoramic video is scored |
CN109165557A (en) * | 2018-07-25 | 2019-01-08 | 曹清 | Scape does not judge system and the other judgment method of scape |
CN110620946A (en) * | 2018-06-20 | 2019-12-27 | 北京优酷科技有限公司 | Subtitle display method and device |
CN111738099A (en) * | 2020-05-30 | 2020-10-02 | 华南理工大学 | Face automatic detection method based on video image scene understanding |
CN112446333A (en) * | 2020-12-01 | 2021-03-05 | 中科人工智能创新技术研究院(青岛)有限公司 | Ball target tracking method and system based on re-detection |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN100531352C (en) * | 2007-08-10 | 2009-08-19 | 西安交通大学 | A method for detecting wonderful section of football match video |
CN101324957B (en) * | 2008-07-16 | 2011-06-08 | 上海大学 | Intelligent playing method of football video facing to mobile equipment |
-
2009
- 2009-07-17 CN CN 200910089438 patent/CN101645137B/en not_active Expired - Fee Related
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101794451A (en) * | 2010-03-12 | 2010-08-04 | 上海交通大学 | Tracing method based on motion track |
CN102005053A (en) * | 2010-11-17 | 2011-04-06 | 中国科学院声学研究所 | Player position automatic detection method |
CN104778459A (en) * | 2015-04-17 | 2015-07-15 | 广西科技大学 | Feature fusion method for actions of multiple athletes in football match video |
CN104794488A (en) * | 2015-04-17 | 2015-07-22 | 广西科技大学 | Transfer learning method of team performance classification indicated by sparse prototype |
CN104866853A (en) * | 2015-04-17 | 2015-08-26 | 广西科技大学 | Method for extracting behavior characteristics of multiple athletes in football match video |
CN106991358A (en) * | 2016-01-20 | 2017-07-28 | 上海慧体网络科技有限公司 | The algorithm that automatic identification football match based on panoramic video is scored |
CN106651952A (en) * | 2016-10-27 | 2017-05-10 | 深圳锐取信息技术股份有限公司 | Football detecting and tracking based video processing method and device |
CN106651952B (en) * | 2016-10-27 | 2020-10-20 | 深圳锐取信息技术股份有限公司 | Video processing method and device based on football detection and tracking |
CN110620946A (en) * | 2018-06-20 | 2019-12-27 | 北京优酷科技有限公司 | Subtitle display method and device |
CN110620946B (en) * | 2018-06-20 | 2022-03-18 | 阿里巴巴(中国)有限公司 | Subtitle display method and device |
CN109165557A (en) * | 2018-07-25 | 2019-01-08 | 曹清 | Scape does not judge system and the other judgment method of scape |
CN111738099A (en) * | 2020-05-30 | 2020-10-02 | 华南理工大学 | Face automatic detection method based on video image scene understanding |
CN111738099B (en) * | 2020-05-30 | 2023-11-07 | 华南理工大学 | Face automatic detection method based on video image scene understanding |
CN112446333A (en) * | 2020-12-01 | 2021-03-05 | 中科人工智能创新技术研究院(青岛)有限公司 | Ball target tracking method and system based on re-detection |
CN112446333B (en) * | 2020-12-01 | 2023-05-02 | 中科人工智能创新技术研究院(青岛)有限公司 | Ball target tracking method and system based on re-detection |
Also Published As
Publication number | Publication date |
---|---|
CN101645137B (en) | 2012-07-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN101645137B (en) | Method for automatically detecting location of a football in long shot of football video | |
Chiu et al. | A robust object segmentation system using a probability-based background extraction algorithm | |
CN102201146B (en) | Active infrared video based fire smoke detection method in zero-illumination environment | |
CN103258232B (en) | A kind of public place crowd estimate's method based on dual camera | |
US20120288189A1 (en) | Image processing method and image processing device | |
CN102622584B (en) | Method for detecting mask faces in video monitor | |
CN104978567B (en) | Vehicle checking method based on scene classification | |
CN102084397B (en) | Image processing device and method | |
CN101739561A (en) | TV station logo training method and identification method | |
CN103425967A (en) | Pedestrian flow monitoring method based on pedestrian detection and tracking | |
CN102156880A (en) | Method for detecting abnormal crowd behavior based on improved social force model | |
CN103605991A (en) | Automatic video advertisement detection method | |
CN103226860B (en) | Passage passenger traffic density estimation method | |
CN109145708A (en) | A kind of people flow rate statistical method based on the fusion of RGB and D information | |
CN104102910A (en) | Sports video tactical behavior recognition method based on space-time local mode | |
CN101324957B (en) | Intelligent playing method of football video facing to mobile equipment | |
CN104463232A (en) | Density crowd counting method based on HOG characteristic and color histogram characteristic | |
CN107483894A (en) | Judge to realize the high ferro station video monitoring system of passenger transportation management based on scene | |
CN108471497A (en) | A kind of ship target real-time detection method based on monopod video camera | |
Wang et al. | Early smoke detection in video using swaying and diffusion feature | |
CN115661720A (en) | Target tracking and identifying method and system for shielded vehicle | |
CN102867214B (en) | Counting management method for people within area range | |
CN101877135B (en) | Moving target detecting method based on background reconstruction | |
CN105989615A (en) | Pedestrian tracking method based on multi-feature fusion | |
CN105118073A (en) | Human body head target identification method based on Xtion camera |
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: 20120704 Termination date: 20150717 |
|
EXPY | Termination of patent right or utility model |