CN102148919B - Method and system for detecting balls - Google Patents
Method and system for detecting balls Download PDFInfo
- Publication number
- CN102148919B CN102148919B CN201010110063.4A CN201010110063A CN102148919B CN 102148919 B CN102148919 B CN 102148919B CN 201010110063 A CN201010110063 A CN 201010110063A CN 102148919 B CN102148919 B CN 102148919B
- Authority
- CN
- China
- Prior art keywords
- color
- area
- candidate
- interest
- ball
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a method and a system for detecting balls. The method comprises the following steps: extracting reference color information of a target ball from a ball image sample; measuring the difference between the reference color of the target ball and the color of an image to be detected by using a preset coarse search color measurement threshold, and determining a candidate interest region in the image to be detected according to the color difference measurement result, wherein the candidate interest region comprises cluster factors; measuring the difference between the reference color of the target ball and the colors of candidate interest regions in the image to be detected by using a preset precise search color measurement threshold, re-determining the cluster factors of the candidate interest regions in the image to be detected on the basis of the color difference measurement result; extracting the characteristics of the target ball of the candidate interest region, matching the characteristic value of the target ball with a characteristic reference value so as to obtain a target ball region. The method and the system are applicable to a static camera shooting mode and further applicable to a movement camera shooting mode and can effectively improve the detection precision of ball targets.
Description
Technical field
The present invention relates to the technical field of computer vision, particularly relate to a kind of method of ball detection and a kind of system of ball detection.
Background technology
Sports Video Analysis research based on multimedia technology serves primarily in two large application directions: one of them is content-based analysis and searching system, and the target detection technique based on the picture taken under motion cameras is one of key technology of this system; Two is computer vision systems of assistant judge based on video monitoring and technical-tactics analyzing.
In prior art, for the study general of the computer vision system of ball game based on tennis " instant playback system ", the tennis vision system as hawk-eye is just mentioned and is carried out detection tennis with simple image treatment technology; The tennis vision system of another kind of Lucentvision be first combine the region utilizing front and back frame subtract to extract motion, and the result of both prediction gray values "AND" ball is in the current frame as the candidate region of target; Then utilize simple ball aspect ratio such as size, shape (length-width ratio) to carry out filtering, finally obtain detecting target.
But, the ball detection method of this prior art, because it adopts simple image analysis technology, or frame difference extracts candidate target before and after adopting, default gray value is utilized to locate candidate target at present frame, location is not very accurate, the image of motion cameras shooting cannot be applicable in addition, especially for the ball detection of table tennis, because sportsman's table tennis when serving a ball can be placed in athletic palm, movable information is not had when ball is placed on ball in palm, the ball detection method of prior art is adopted also to there will be the problem that ball cannot be detected completely.
Therefore, the technical problem needing those skilled in the art urgently to solve at present is exactly: how innovatively can propose a kind of ball testing mechanism, with on the basis being applicable to Still Camera screening-mode, be applicable to motion cameras screening-mode further, and effectively improve the accuracy of detection of ball target.
Summary of the invention
Technical problem to be solved by this invention is to provide a kind of method and system of ball detection, with on the basis being applicable to Still Camera screening-mode, is applicable to motion cameras screening-mode further, and effectively improves the accuracy of detection of ball target.
In order to solve the problems of the technologies described above, the embodiment of the invention discloses a kind of method of ball detection, comprising:
The reference color information of object ball is extracted from ball image pattern;
Adopt preset coarse search color measurements threshold value, measure the reference color of described object ball and the color distance of image to be detected, and in image to be detected, determine candidate's area-of-interest according to described color distance measurement results, described candidate's area-of-interest comprises clustering factor;
Adopt preset essence search color measurements threshold value, measure the color distance of each candidate's area-of-interest in the reference color of described object ball and image to be detected, and in image to be detected, redefine the clustering factor in described candidate's area-of-interest according to described color distance measurement results;
Extract the object ball feature of described candidate's area-of-interest, and described object ball characteristic value is mated with feature reference value, obtain object ball region.
Preferably, the described step extracting the reference color information of object ball from image pattern comprises further:
Each Color Channel for ball image pattern builds one dimension average color histogram;
Determine the peak value index of each one dimension average histogram;
According to described peak value index interval of definition, the border in described interval is less than the pixel value of peak value index;
According to the average color in each Color Channel of described interval computation, as the reference color information of object ball.
Preferably, described foundation color distance measurement results determines that in image to be detected the step of candidate's area-of-interest comprises further:
Adopt described color distance measurement results to travel through pixel in image to be detected, obtain the pixel meeting threshold range as clustering factor;
Described clustering factor according to its in image to be detected position composition cluster areas blob;
Size and location information according to described cluster areas blob determines candidate's area-of-interest.
Preferably, the step of the described clustering factor redefined in image to be detected in described candidate's area-of-interest according to color distance measurement results comprises further:
Adopt described color distance measurement results to travel through pixel in image to be detected in candidate's area-of-interest, obtain the pixel meeting threshold range as the clustering factor redefined.
Preferably, described object ball feature comprises the symmetry information of area information, region long axis information, region minor axis information, all long messages, information on eccentricity, circularity information and circle.
Preferably, describedly object ball characteristic value and feature reference value carried out mating the step obtaining object ball region comprise further:
If described object ball characteristic value all meets corresponding feature reference value, then judge that current candidate area-of-interest is as object ball region;
And/or,
If the symmetry information of described area information, information on eccentricity, circularity information and circle meets corresponding feature reference value, then judge that current candidate area-of-interest is as object ball region;
And/or,
If a two field picture occurs plural candidate's area-of-interest, then judge that the maximum candidate's area-of-interest of circularities is as object ball region.
Preferably, described method, also comprises:
The detection of an object ball is started at start frame or every N frame.
The embodiment of the invention also discloses a kind of system of ball detection, comprising:
Reference color statistical module, for extracting the reference color information of object ball from ball image pattern;
Coarse search module, for adopting preset coarse search color measurements threshold value, measure the reference color of described object ball and the color distance of image to be detected, and in image to be detected, determine candidate's area-of-interest according to described color distance measurement results, described candidate's area-of-interest comprises clustering factor;
Essence search module, for adopting preset essence search color measurements threshold value, measure the color distance of each candidate's area-of-interest in the reference color of described object ball and image to be detected, and in image to be detected, redefine the clustering factor in described candidate's area-of-interest according to described color distance measurement results;
Target verification module, for extracting the object ball feature of described candidate's area-of-interest, and mates described object ball characteristic value with feature reference value, obtains object ball region.
Preferably, described reference color statistical module comprises further:
Histogram builds submodule, builds one dimension average color histogram for each Color Channel for ball image pattern;
Peak value index determination submodule, for determining the peak value index of each one dimension average histogram;
Section definition submodule, for according to described peak value index interval of definition, the border in described interval is less than the pixel value of peak value index;
Reference color calculating sub module, for according to the average color in each Color Channel of described interval computation, as the reference color information of object ball.
Preferably, described coarse search module comprises further:
Pixel traversal submodule, for adopting described color distance measurement results to travel through pixel in image to be detected, obtains the pixel meeting threshold range as clustering factor;
Cluster submodule, for described clustering factor according to its in image to be detected position composition cluster areas blob;
ROI determines submodule, determines candidate's area-of-interest for the size and location information according to described cluster areas blob.
Preferably, described smart search module comprises further:
The accurate locator module of target, for adopting described color distance measurement results to travel through pixel in image to be detected in candidate's area-of-interest, obtains the pixel meeting threshold range as the clustering factor redefined.
Preferably, described object ball feature comprises the symmetry information of area information, region long axis information, region minor axis information, all long messages, information on eccentricity, circularity information and circle.
Preferably, described target verification module comprises further:
First syndrome module, for when described object ball characteristic value all meets corresponding feature reference value, judges that current candidate area-of-interest is as object ball region;
And/or,
Second syndrome module, for when the symmetry information of described area information, information on eccentricity, circularity information and circle meets corresponding feature reference value, judges that current candidate area-of-interest is as object ball region;
And/or,
3rd syndrome module, for there is plural candidate's area-of-interest on a two field picture, judges that the maximum candidate's area-of-interest of circularities is as object ball region.
Compared with prior art, the present invention has the following advantages:
First, the present invention adopts by the thick method to essence search area-of-interest, is extracted the area-of-interest that all may contain target by coarse search, then may carry out essence search containing in the area-of-interest of target, to make the details display of target completely, so the precision of target localization is high; Further, area-of-interest determines according to the size adaptation of target, thus can surround target rightly and be unlikely to again to surround much more irrelevant regions.
Secondly, the present invention adopts associating area, eccentricity, circularity, the ball feature evaluation target of symmetry four kinds, can more effectively identify ball target, thus effectively improve the accuracy of detection of ball target.
Moreover the reference color of the present invention's all right online adaptive adjustment ball, makes it the change adapting to illumination, the distance more accurately between metric objective and reference color, and make color of object cluster more complete, it is more accurate to locate.
Finally, the present invention can also add up the reference color of ball, makes detection algorithm have the statistical property learning ball obvious color, makes these statistics adapt to the change of image-forming condition, and the ball of different sports items can be detected, with the ball of what color in no matter competing.
In addition, because the present invention is not easy by illumination, background influence, the scope of application is wide, can also process the ball be placed on when sportsman serves a ball in palm further, and during ball rapid movement, imaging is not the ball justified especially.
Accompanying drawing explanation
Fig. 1 is the flow chart of the embodiment of the method for a kind of ball detection of the present invention;
Fig. 2 a is the original image of the frame ping-pong contest picture as the ball detection example of the present invention;
Fig. 2 b is the display schematic diagram obtaining cluster blob after carrying out coarse search to the image shown in Fig. 2 a;
Fig. 2 c is the display schematic diagram determining ROI based on the cluster blob self adaptation shown in Fig. 2 b;
Fig. 2 d be based on the further localizing objects of ROI shown in Fig. 2 c after display schematic diagram;
Fig. 2 e is the display schematic diagram of the final detection objective result obtained after carrying out goal-based assessment to the ROI shown in Fig. 2 d;
Fig. 3 is the computational methods schematic diagram that a kind of self adaptation of the present invention determines ROI;
Fig. 4 is the structured flowchart of the system embodiment of a kind of ball detection of the present invention.
Embodiment
For enabling above-mentioned purpose of the present invention, feature and advantage become apparent more, and below in conjunction with the drawings and specific embodiments, the present invention is further detailed explanation.
With reference to figure 1, show the flow chart of the embodiment of the method for a kind of ball detection of the present invention, specifically can comprise the following steps:
Step 101, from ball image pattern, extract the reference color information of object ball;
Step 102, adopt preset coarse search color measurements threshold value, measure the reference color of described object ball and the color distance of image to be detected, and in image to be detected, determine candidate's area-of-interest according to described color distance measurement results, described candidate's area-of-interest comprises clustering factor;
Step 103, adopt preset essence search color measurements threshold value, measure the color distance of each candidate's area-of-interest in the reference color of described object ball and image to be detected, and in image to be detected, redefine the clustering factor in described candidate's area-of-interest according to described color distance measurement results;
Step 104, extract the object ball feature of described candidate's area-of-interest, and described object ball characteristic value is mated with feature reference value, obtain object ball region.
For making those skilled in the art understand the present invention better, the visual analysis system of ball game in relaying for competitive sports below, key technology and principle simple declaration to ball detection:
For the ball detection technique of the plane of delineation.The target detection scheme of view-based access control model is generally divided into two basic steps: one is hypothesis generation stage (Hypothesis Generation, HG), also candidate region is made to determine the stage, through Iamge Segmentation, can produce and may comprise potential ball area-of-interest (Region Of Interest, ROI), and accurately locate potential ball target position in the roi, namely suppose potential target location in the picture; Two is hypothesis the stage of recognition (HypothesisVerification, HV), namely confirms the candidate region produced on last stage, judges whether it is target, thus detect target.
Can learn in conjunction with above-mentioned principle, the step 103 of step 101-described in the embodiment of the present invention is the hypothesis generation stage, and step 104 is hypothesis the stage of recognition.The present invention adopts by the thick ball candidate region detection method to essence (coarse-to-fine) in the hypothesis generation stage, the ball features such as size, eccentricity, circularity, symmetry are adopted to carry out filtering at hypothesis the stage of recognition, namely this ball detection method is applicable to the image of Still Camera shooting, is also applicable to the image of motion cameras shooting.
In one preferred embodiment of the invention, described step 101 may further include following sub-step:
Sub-step A1, build one dimension average color histogram for each Color Channel of ball image pattern;
Sub-step A2, determine the peak value index of each one dimension average histogram;
Sub-step A3, according to described peak value index interval of definition, the border in described interval is less than the pixel value of peak value index;
Sub-step A4, according to the average color in each Color Channel of described interval computation, as the reference color information of object ball.
Namely in embodiments of the present invention, the method for statistics specifically can be adopted to estimate the reference color of color as object ball of typical ball.For the ball picture sample set that will extract color, each Color Channel uses all pixels to build one dimension average color histogram, and determines the peak value index of each one dimension average histogram, near peak value index, define an interval [i
min, i
max], the pixel of interval border n% fewer than peak value index, wherein, n can get arbitrary value by those skilled in the art depending on concrete condition.After determining interval border, the average color that just can calculate in interval for each Color Channel is:
Minitial＝[r
M，g
M，b
M]
Wherein, r
mg
mb
mas currency or initial value.
In practice, by the method for color detection ball be exactly the most simply: utilize the threshold value that the reference color of object ball goes estimation one fixing.But, because the color of ball is subject to illumination effect not to be uniform usually; In addition, in different positions, the brightness of ball is changeable; Further, in difference match scene, the brightness ratio average of scene is bright or dark.Obviously this fixed threshold estimated by reference color is no longer suitable, and when transformation diagram picture is to bianry image, (a lot of independent object occurs will to cause over-segmentation (ball portion loss) or weak segmentation.)
In order to address this problem, the present invention proposes a kind of searching method by the thick two benches threshold value to essence.Adopt in this way, in specific implementation, need first to generate coarse search color measurements threshold value and essence search color measurements threshold value according to the reference color information of object ball, these two threshold values can adopt heuristic to estimate in practice, because embody rule scene exists each species diversity, therefore the present invention does not require that these two threshold restrictions are in a certain scope.
Specifically, in the coarse search stage, can adopt the threshold value that a scope is less, as the arbitrary value between 30 1-50, object is the candidate target having Similar color in order to only leave those and ball, and remaining is all deleted.
In one preferred embodiment of the invention, described step 102 specifically can comprise following sub-step:
Sub-step B 1, adopt described color distance measurement results to travel through pixel in image to be detected, obtain the pixel meeting threshold range as clustering factor;
As a kind of example, the method that described traversal pixel carries out color distance tolerance can adopt euclidean metric (L
2norm), specifically can pass through following formulae discovery:
Wherein, x
jand x
kfor color vector, be suitable for all color spaces except HSI, j and k is jth and a kth pixel respectively, and p represents Color Channel number, and at this, p is that 2, l represents each Color Channel, is respectively r and g Color Channel.
Sub-step B2, described clustering factor according to its in image to be detected position composition cluster areas blob;
Sub-step B3, determine candidate's area-of-interest according to the size and location information of described cluster areas blob.
Further illustrate the process of above-mentioned coarse search below in conjunction with the ping-pong contest picture shown in Fig. 2, wherein, the original image of this ping-pong contest picture as shown in Figure 2 a.
As shown in Figure 2 b, meet the pixel of threshold range with object ball reference color distance, all stayed in bianry image, and be clustered into N number of potential ball region blob.For each potential ball region adaptivity determination area-of-interest (ROI).The method determined as shown in Figure 3, namely for the blob of cluster, first obtain the radius Wradius of center position and the radius Hradius in transverse and longitudinal direction, and, the border of its surrounding, then in the boundary rectangle surrounding of this blob, then respectively to outer expansion a*radius, wherein, described a can arrange arbitrary value or employing empirical value by those skilled in the art according to actual conditions.Adopt this size according to blob self adaptation can determine ROI, and this ROI determined can surround ball candidate target completely, be unlikely to again to surround too large irrelevant region.Simultaneously for these ROI produced, preliminary filtering can also be carried out according to area and aspect ratio, not being that ball ROI candidate filters completely, finally produce some potential ball ROIs as shown in Figure 2 c.
Because many irrelevant targets have just been removed in the coarse search stage, so the processing time of analysis image needs can be reduced, and, after being converted to bianry image, there is the chance that the target similar with ball occurs.
In the essence search phase, for the ROI region of each candidate, more wide in range threshold value can be used, as got arbitrary value between 50--80.Certainly, above-mentioned span is only used as example.Remaining target is owing to using wide in range threshold value in ROI, and ball details can be displayed completely out, thus reaches the object of accurate localizing objects.
In one preferred embodiment of the invention, the process of described step 103 is specifically as follows, adopt described color distance measurement results, travel through the pixel in candidate's area-of-interest in image to be detected, obtain the pixel meeting threshold range as the clustering factor redefined.After the essence search phase, the cluster blob in corresponding ROI namely as shown in Figure 2 d.
After target is accurately located, just can have evaluated the candidate target detected (the candidate's area-of-interest through the essence search phase).In order to verify whether detected target is object ball, needing to calculate corresponding object ball feature, and carrying out assessing and comparing with preset feature reference value.As a kind of example, described object ball feature and account form specifically can with reference to following tables:
Ideally, when the object ball characteristic sum feature reference value of candidate's area-of-interest matches, so target just can be identified as ball target.But due to the factor such as ambient interferences of illumination difference and class ball color, in most of the cases some features can not match.In addition, during service, ball is in the palm of serving side, and ball is not round, because its bottom may be covered by sportsman's palm.Thus, in order to can target be detected, in specific implementation, can setting area, eccentricity, circularity, symmetric characteristics feature reference value as shown in the table:
| feature | rule |
| area (A) | according to the resolution of image, an interval threshold [Amin Amax] is suitably set |
| eccentricity (E) | be less than 1.4 |
| roundness (R) | be greater than 0.3 |
| symmetry (S) | 12 symmetry (Sym1-sym12) are less than a wide in range threshold value r/n, illustrate target be have symmetric. |
If all meet these conditions simultaneously, just illustrate that this target is ball target; If when a two field picture occurring two targets ball with two or more, then compare circularity, maximum that target of circularities is detected ball target.
Such as, search for obtain 7 ball target ROI of candidate according to the ROIs in Fig. 2 c through essence in Fig. 2 d, for these seven ball target ROI of candidate, calculate by above-mentioned feature calculation method, then utilize above-mentioned feature reference value to judge, thus draw assessment result as shown in the table:
As can be seen from the above table, owing to only having candidate target 1 all to meet 4 conditions, so the last ball target detected is exactly candidate target 1, as shown in the rectangle frame at the table tennis position in Fig. 4 e.
In specific implementation, the reference color of all right self-adaptative adjustment ball of the embodiment of the present invention.Specifically, after detected ball target is assessed, its circularity Roundness belongs to that [0.91] is interval, eccentricity is less than 1.2, area and symmetry all within certain scope in, this target is put into buffer memory buffer, this condition is all met when running up to five frame targets, the then reference color information of on-line tuning object ball, to adapt to the change of illumination.
Such as, the average of this five frame is got as Mlocal (Mlocal=[r
m, g
m, b
m]) value, the reference color of the ball after renewal is Mcurr (Mcurr=[r
m, g
m, b
m]).；
Corresponding more new formula is: Mcurr=w
l* Mlocal+w
i* Minitial
Wherein w
i≈ 0.5 ~ 1.0, w
l≈ 1-w
i
Adopt the present invention to utilize the picture in real racetrack scene to carry out ball detection, under background complicated situation, still can obtain good Detection results.
It should be noted that, the present invention is mainly applicable to clap the ball detection of the rule of type games, and as tennis, table tennis etc., ball for other rules, the such as detection of football, basketball, vollyball also has important reference value.
For aforesaid each embodiment of the method, in order to simple description, therefore it is all expressed as a series of combination of actions, but those skilled in the art should know, the present invention is not by the restriction of described sequence of movement, because according to the present invention, some step can adopt other orders or carry out simultaneously.Secondly, those skilled in the art also should know, the embodiment described in specification all belongs to preferred embodiment, and involved action and module might not be that the present invention is necessary.
With reference to figure 4, show the structured flowchart of the system embodiment of a kind of ball detection of the present invention, specifically can comprise with lower module:
Reference color statistical module 401, for extracting the reference color information of object ball from ball image pattern;
Coarse search module 402, for adopting preset coarse search color measurements threshold value, measure the reference color of described object ball and the color distance of image to be detected, and in image to be detected, determine candidate's area-of-interest according to described color distance measurement results, described candidate's area-of-interest comprises clustering factor;
Essence search module 403, for adopting preset essence search color measurements threshold value, measure the color distance of each candidate's area-of-interest in the reference color of described object ball and image to be detected, and in image to be detected, redefine the clustering factor in described candidate's area-of-interest according to described color distance measurement results;
Target verification module 404, for extracting the object ball feature of described candidate's area-of-interest, and mates described object ball characteristic value with feature reference value, obtains object ball region.
In embodiments of the present invention, preferably, described reference color statistical module 401 comprises following submodule further:
Histogram builds submodule, builds one dimension average color histogram for each Color Channel for ball image pattern;
Peak value index determination submodule, for determining the peak value index of each one dimension average histogram;
Section definition submodule, for according to described peak value index interval of definition, the border in described interval is less than the pixel value of peak value index;
Reference color calculating sub module, for according to the average color in each Color Channel of described interval computation, as the reference color information of object ball.
In embodiments of the present invention, preferably, described coarse search module 402 may further include following submodule:
Pixel traversal submodule, for adopting described color distance measurement results to travel through pixel in image to be detected, obtains the pixel meeting threshold range as clustering factor;
Cluster submodule, for described clustering factor according to its in image to be detected position composition cluster areas blob;
ROI determines submodule, determines candidate's area-of-interest for the size and location information according to described cluster areas blob.
In embodiments of the present invention, preferably, described smart search module 403 may further include following submodule:
The accurate locator module of target, for adopting described color distance measurement results to travel through pixel in image to be detected in candidate's area-of-interest, obtains the pixel meeting threshold range as the clustering factor redefined.
In specific implementation, described object ball feature can comprise the symmetry information of area information, region long axis information, region minor axis information, all long messages, information on eccentricity, circularity information and circle.
In this case, described target verification module 404 may further include following submodule:
First syndrome module, for when described object ball characteristic value all meets corresponding feature reference value, judges that current candidate area-of-interest is as object ball region;
And/or,
Second syndrome module, for when the symmetry information of described area information, information on eccentricity, circularity information and circle meets corresponding feature reference value, judges that current candidate area-of-interest is as object ball region;
And/or,
3rd syndrome module, for there is plural candidate's area-of-interest on a two field picture, judges that the maximum candidate's area-of-interest of circularities is as object ball region.
Because system embodiment of the present invention is substantially corresponding to aforesaid embodiment of the method, therefore not detailed part in the description of the present embodiment, see the related description in previous embodiment, just can not repeat at this.
The present invention can be used in numerous general or special purpose computing system environment or configuration.Such as: multicomputer system, server, network PC, minicom, mainframe computer, the distributed computing environment (DCE) comprising above any system or equipment etc.
The present invention can describe in the general context of computer executable instructions, such as program module.Usually, program module comprises the routine, program, object, assembly, data structure etc. that perform particular task or realize particular abstract data type.Also can put into practice the present invention in a distributed computing environment, in these distributed computing environment (DCE), be executed the task by the remote processing devices be connected by communication network.In a distributed computing environment, program module can be arranged in the local and remote computer-readable storage medium comprising memory device.
Above the method for a kind of ball detection provided by the present invention and a kind of system of ball detection are described in detail, apply specific case herein to set forth principle of the present invention and execution mode, the explanation of above embodiment just understands method of the present invention and core concept thereof for helping; Meanwhile, for one of ordinary skill in the art, according to thought of the present invention, all will change in specific embodiments and applications, in sum, this description should not be construed as limitation of the present invention.
Claims (3)
1. a method for ball detection, is characterized in that, comprising:
The reference color information of object ball is extracted from ball image pattern;
Adopt preset coarse search color measurements threshold value, measure the reference color of described object ball and the color distance of image to be detected, and in image to be detected, determine candidate's area-of-interest according to described color distance measurement results, described candidate's area-of-interest comprises clustering factor;
Adopt preset essence search color measurements threshold value, measure the color distance of each candidate's area-of-interest in the reference color of described object ball and image to be detected, and in image to be detected, redefine the clustering factor in described candidate's area-of-interest according to described color distance measurement results;
The object ball feature of the candidate's area-of-interest redefined described in extraction, and described object ball characteristic value is mated with feature reference value, obtain object ball region;
Wherein, the described step extracting the reference color information of object ball from image pattern comprises further:
Each Color Channel for ball image pattern builds one dimension average color histogram;
Determine the peak value index of each one dimension average histogram;
According to described peak value index interval of definition, the border in described interval is less than the pixel value of peak value index;
According to the average color in each Color Channel of described interval computation, as the reference color information of object ball;
Described foundation color distance measurement results determines that in image to be detected the step of candidate's area-of-interest comprises further:
Adopt described color distance measurement results to travel through pixel in image to be detected, obtain the pixel meeting threshold range as clustering factor;
Described clustering factor according to its in image to be detected position composition cluster areas blob;
Size and location information according to described cluster areas blob determines candidate's area-of-interest;
The step of the clustering factor that described foundation color distance measurement results redefines in described candidate's area-of-interest in image to be detected comprises further:
Adopt described color distance measurement results to travel through pixel in image to be detected in candidate's area-of-interest, obtain the pixel meeting threshold range as the clustering factor redefined;
Described object ball feature comprises the symmetry information of area information, region long axis information, region minor axis information, all long messages, information on eccentricity, circularity information and circle;
Describedly object ball characteristic value and feature reference value carried out mating the step obtaining object ball region comprise further:
If described object ball characteristic value all meets corresponding feature reference value, then judge that current candidate area-of-interest is as object ball region;
And/or,
If the symmetry information of described area information, information on eccentricity, circularity information and circle meets corresponding feature reference value, then judge that current candidate area-of-interest is as object ball region;
And/or,
If a two field picture occurs plural candidate's area-of-interest, then judge that the maximum candidate's area-of-interest of circularities is as object ball region.
2. the method for claim 1, is characterized in that, also comprises:
The detection of an object ball is started at start frame or every N frame.
3. a system for ball detection, is characterized in that, comprising:
Reference color statistical module, for extracting the reference color information of object ball from ball image pattern;
Coarse search module, for adopting preset coarse search color measurements threshold value, measure the reference color of described object ball and the color distance of image to be detected, and in image to be detected, determine candidate's area-of-interest according to described color distance measurement results, described candidate's area-of-interest comprises clustering factor;
Essence search module, for adopting preset essence search color measurements threshold value, measure the color distance of each candidate's area-of-interest in the reference color of described object ball and image to be detected, and in image to be detected, redefine the clustering factor in described candidate's area-of-interest according to described color distance measurement results;
Target verification module, for the object ball feature of candidate's area-of-interest redefined described in extracting, and mates described object ball characteristic value with feature reference value, obtains object ball region;
Wherein, described reference color statistical module comprises further:
Histogram builds submodule, builds one dimension average color histogram for each Color Channel for ball image pattern;
Peak value index determination submodule, for determining the peak value index of each one dimension average histogram;
Section definition submodule, for according to described peak value index interval of definition, the border in described interval is less than the pixel value of peak value index;
Reference color calculating sub module, for according to the average color in each Color Channel of described interval computation, as the reference color information of object ball;
Described coarse search module comprises further:
Pixel traversal submodule, for adopting described color distance measurement results to travel through pixel in image to be detected, obtains the pixel meeting threshold range as clustering factor;
Cluster submodule, for described clustering factor according to its in image to be detected position composition cluster areas blob;
ROI determines submodule, determines candidate's area-of-interest for the size and location information according to described cluster areas blob;
Described smart search module comprises further:
The accurate locator module of target, for adopting described color distance measurement results to travel through pixel in image to be detected in candidate's area-of-interest, obtains the pixel meeting threshold range as the clustering factor redefined;
Described object ball feature comprises the symmetry information of area information, region long axis information, region minor axis information, all long messages, information on eccentricity, circularity information and circle;
Described target verification module comprises further:
First syndrome module, for when described object ball characteristic value all meets corresponding feature reference value, judges that current candidate area-of-interest is as object ball region;
And/or,
Second syndrome module, for when the symmetry information of described area information, information on eccentricity, circularity information and circle meets corresponding feature reference value, judges that current candidate area-of-interest is as object ball region;
And/or,
3rd syndrome module, for there is plural candidate's area-of-interest on a two field picture, judges that the maximum candidate's area-of-interest of circularities is as object ball region.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201010110063.4A CN102148919B (en) | 2010-02-09 | 2010-02-09 | Method and system for detecting balls |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201010110063.4A CN102148919B (en) | 2010-02-09 | 2010-02-09 | Method and system for detecting balls |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102148919A CN102148919A (en) | 2011-08-10 |
| CN102148919B true CN102148919B (en) | 2015-05-27 |
Family
ID=44422894
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201010110063.4A Expired - Fee Related CN102148919B (en) | 2010-02-09 | 2010-02-09 | Method and system for detecting balls |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102148919B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103886618A (en) * | 2014-03-12 | 2014-06-25 | 新奥特（北京）视频技术有限公司 | Ball detection method and device |
| CN110290426B (en) * | 2019-06-24 | 2022-04-19 | 腾讯科技（深圳）有限公司 | Method, device and equipment for displaying resources and storage medium |
| CN110728659A (en) * | 2019-09-17 | 2020-01-24 | 深圳新视智科技术有限公司 | Defect merging method and device, computer equipment and storage medium |
| WO2021100516A1 (en) * | 2019-11-22 | 2021-05-27 | ソニーグループ株式会社 | Information processing device, information processing method, and program |
| CN111539910B (en) * | 2020-03-13 | 2024-03-19 | 石家庄铁道大学 | Rust area detection method and terminal equipment |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1905629A (en) * | 2005-07-26 | 2007-01-31 | 佳能株式会社 | Image capturing apparatus and image capturing method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6741725B2 (en) * | 1999-05-26 | 2004-05-25 | Princeton Video Image, Inc. | Motion tracking using image-texture templates |
-
2010
- 2010-02-09 CN CN201010110063.4A patent/CN102148919B/en not_active Expired - Fee Related
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1905629A (en) * | 2005-07-26 | 2007-01-31 | 佳能株式会社 | Image capturing apparatus and image capturing method |
Non-Patent Citations (1)
| Title |
|---|
| 孙小林,王建宇.体育视频中运动目标的检测与跟踪.《南京理工大学学位论文》.2009, * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102148919A (en) | 2011-08-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112861720B (en) | Remote sensing image small sample target detection method based on prototype convolutional neural network | |
| CN109544592B (en) | Moving object detection algorithm for camera movement | |
| CN109145708B (en) | Pedestrian flow statistical method based on RGB and D information fusion | |
| CN109903312A (en) | A kind of football sportsman based on video multi-target tracking runs distance statistics method | |
| CN111368791B (en) | Pull-up test counting method and system based on Quick-OpenPose model | |
| CN108765470A (en) | One kind being directed to the improved KCF track algorithms of target occlusion | |
| CN102148919B (en) | Method and system for detecting balls | |
| CN101770583B (en) | Template matching method based on global features of scene | |
| Liu et al. | Extracting 3D information from broadcast soccer video | |
| CN103413149B (en) | Method for detecting and identifying static target in complicated background | |
| CN108492298A (en) | Based on the multispectral image change detecting method for generating confrontation network | |
| CN110084830A (en) | A kind of detection of video frequency motion target and tracking | |
| CN113744316A (en) | Multi-target tracking method based on deep neural network | |
| CN108765463A (en) | A kind of moving target detecting method calmodulin binding domain CaM extraction and improve textural characteristics | |
| Karavasilis et al. | Visual tracking by adaptive Kalman filtering and mean shift | |
| Liu et al. | Research on action recognition of player in broadcast sports video | |
| CN106446837B (en) | A kind of detection method of waving based on motion history image | |
| CN103903258B (en) | Method for detecting change of remote sensing image based on order statistic spectral clustering | |
| Leong et al. | Computer vision approach to automatic linesman | |
| CN114612934A (en) | Gait sequence evaluation method and system based on quality dimension | |
| CN114463664A (en) | Novel ice hockey tracking method for ice hockey sports | |
| Cruz et al. | Deep learning applied to humanoid soccer robotics: playing without using any color information | |
| Spagnolo et al. | On-field testing and evaluation of a goal-line technology system | |
| CN112417961A (en) | Sea surface target detection method based on scene prior knowledge | |
| Chi et al. | Concrete Application of Computer Virtual Image Technology in Modern Sports Training |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20150527 Termination date: 20170209 |