CN101833232B - Visual support and match analysis system for ping-pong match and method for running same - Google Patents

Abstract

The invention provides a visual support and match analysis system for a ping-pong match. The system comprises the following modules: an image acquiring and ping-pong ball target recognizing and positioning module, a KALMAN filter module, a track data fitter module, a characteristic analyzing module, a kinematic parameter learning machine module, a single path judging module, a three-dimensional virtual scene reappearing module, a match characteristic data storing module and a user interaction module. A set of system with friendly and complete information extraction and storage, user interaction support and capability of performing track recording well and extracting characteristic information on ball hit of an athlete is developed through real-time visual acquisition. The system can process data in real time and filter and record the processed data in ping-pong trainings and matches, automatically extracts ping-pong ball path characteristic, bring convenience to a judge to replay, is applied to automatic judging and television relaying in the ping-pong match, has a complete, effective and overall analyzing function after the match and can automatically extract and store the characteristic information in a match process and intuitively counts and displays ball throwing and ball winning path characteristics of the ping-pong athlete.

Description

The ping-pong contest vision is supported and match analytic system and system operation method

Technical field

The present invention relates to a kind of ping-pong contest vision support and match analytic system, be used for real-time detection, location, tracking, movement locus prediction and the processing of the table tennis flight path of ping-pong contest or training, and the three-dimensional virtual scene that the movement locus and the table tennis drop point information of table tennis are carried out is shown and the historical record playback in real time.

Background technology

" hawkeye " system applies has had the history in 6 or seven years in the judge and the simulation demo of tennis race, has obtained promoting widely in tennis is relayed, and very positive effect has been played in the research of tennis and the fairness of tennis tournament.The formal name of " hawkeye " system is called " instant playback system ", and this system is made up of 8 or 10 high-speed camera heads, four computers and giant-screen." hawkeye " image data amount that system collected is very huge, calculates quite complicatedly, and it demonstrated time that whole process consumes about 10 seconds from the data acquisition to result, do not possess hard real-time.In regulation game, guarantee the smoothness of match and continuously the fairness and the sight of competing had very important meaning, " hawkeye " can only be as match judge's supplementary means, and the number of times of every innings of match challenge of sportsman hawkeye also is conditional.Therefore " hawkeye " system more can only be the processing afterwards that is used for television relay at present, and can not accomplish online judgement.And this system cost cost is very high, merges in medium and small arena to be not suitable for promoting, and also is not suitable for general physical culture supplemental training.Domestic also have research institution to propose small-sized hawkeye system based on less camera (2), but the real-time problem of system does not still solve.

In table tennis, " hawkeye " system of introducing carries out assistant judge and supplemental training, to this, 200910095291.6 patented claim discloses accurate measurement and prediction table tennis track system and system operation method, the real time processing system of can accurately measure, prediction and three-dimensional reappearing the high-speed flight table tennis track.

This system is primarily aimed at the real-time of table tennis image acquisition, finishes calculating, the table tennis flight path of accurate recognition, location and tracking high-speed motion in the extremely short time.The table tennis stroke is very short, and whole effective travel only continues about 0.5s when high-speed motion (3-5m/s), makes accurate detection and Identification table tennis task very difficult.And because table tennis is frivolous relatively, its flight path has been subjected to all multifactor influences such as table tennis spinning, air themperature, damp condition and on-the-spot ventilation more, be difficult to set up a general motion model, brought very big difficulty therefore for generation of table tennis track and prediction.Thereby identifying from image of table tennis, the model of foundation is difficult to guarantee accuracy of predicting.For the deviation reaction and the sensitivity thereof of initial velocity, thereby cause total system prediction location and track processing accuracy to arrive bottleneck.

Secondly, this system sets up a plurality of video cameras according to the needs of place and environment and observes, and is conceived to data acquisition, and in real racetrack and training, operability is not strong; Extract and still do not set up a complete support system for score information, track replay, drop point demonstration and distribution probability thereof, each athletic motion characteristic value that referee and trainer were concerned about.

Therefore, give the very big difficulty of practical operation of table tennis vision system.Wherein the most key problem is to need system can stay certain nargin to the identification error of table tennis under the prerequisite that guarantees real-time; With perfect and friendly judge (coach) member operation system.

Summary of the invention

The objective of the invention is at the in-problem improvement of 200910095291.6 patents, provide the ping-pong contest vision of perfect and friendly judge (coach) member operation system to support and the match analytic system.Be used for real-time detection, location, tracking, movement locus prediction and the processing of the table tennis flight path of ping-pong contest or training, the three-dimensional virtual scene that the movement locus and the table tennis drop point information of table tennis are carried out shows and the historical record playback in real time, and to the analysis and the statistics of table tennis flight path historical data.

The present invention is intended to set up a training backup system, target is: in routine training at ordinary times, can be by data acquisition and analysis, help the trainer to summarize the athletic style of playing and (comprise habitual serving way, rotational speed and direction, the output direction of ball probability, the feature of losing is concluded, drop point statistics or the like); In the process of match, the assistant judge adjudicates, and can relay the specific track of resetting easily, and the recording storage related data is in order to post-game analysis.

To achieve the above object of the invention, the invention provides a kind of high-speed real-time table tennis location, track generation and forecast method and system that utilizes a plurality of video cameras.Use video camera and adopt figure, can carry out real-time follow-up and track record, realize above-mentioned multi-functional table tennis.

Ping-pong contest vision provided by the invention is supported and the match analytic system, two above video cameras, control multiple-cameras are adopted the synchronous device of figure synchronously, and image data acquiring transmitting device, the main control computer that includes the system environments of vision running software, externally connected with display screen are formed; Video camera is fixed on the ping-pong table sky, and visual field intersection covers the effective coverage of table tennis, carries out image acquisition; The synchronous device of image data acquiring transmitting device produces the multichannel periodic pulse signal, connects video camera by control line, and the control multiple-camera is adopted figure synchronously; Video camera links to each other with the image data acquiring card by data line; Data collecting card is by the pci bus of main control computer, and the processor that the data that collect is sent to the image data acquiring transmitting device is handled; Result outputs on the externally connected with display screen of main control computer and shows; Described system mainly comprises following composition function module: image acquisition and table tennis order claim to discern locating module, KALMAN filter module, track data match device module, characteristics analysis module, kinematic parameter learning machine module, one way judge module, three-dimensional virtual scene reproduction module, match characteristic memory module, user interactive module; Described system architecture flow process is: the main control computer epigraph is gathered and table tennis Target Recognition locating module, through Target Recognition, space orientation, the data that generate are in real time sent into the KALMAN filter module carry out filtering, characteristics analysis module is analyzed the movement locus and the current motion state information of table tennis; The one way judge module is judged, under the situation that one way continues, sends into the three-dimensional virtual scene simulation in real time and reappears module simulation reproduction three-dimensional virtual scene; Under the situation that one way finishes, carry out match with KALMAN match device, generate the movement locus feature, the movement locus feature of generation is made a gift to someone and is write down match characteristic memory module record; The message instruction that user interactive module is imported into by the user is controlled the storage and the three-dimensional simulation interface of data in real time.

KALMAN wave filter of the present invention is a kind of high efficiency regressive filter (autoregressive filter), and it can not exclusively and comprise the measurement of noise estimate the state of dynamic system from a series of, dopes the coordinate and the speed of the position of object.

Match characteristic memory module of the present invention is based on STL (Standard Template Library), i.e. STL in the c++ standard library.The main utilization sequence container that STL comprised is set up data structure storage module in the system.Position, the velocity information of control and storing moving track characteristic; The sportsman who extracts in every movement locus the bat point of impingement, rotation, velocity reversal, size, the service method of feature and the track relevant information of winning the match.Because the rule of whole match is complicated, the referee defines has very strong randomness and real-time, and the record of track should not have automatic segmentation intercepting under the situation of interference, propose critical information the referee; (ruling folk prescription foul under the situation that the referee interferes again; Cancelling this must grade), can make things convenient for to such an extent that visit and modification extraction data; Therefore, adopt and in every game, set up a data structure, controlled and store relevant information based on the container of STL.

Three-dimensional virtual scene reproduction module of the present invention is the three-dimensional virtual scene reproduction module based on OpenGL: the table tennis flight path of replaying user appointment in three-dimensional virtual scene; Table tennis drop point and drop point statistics on the ball table.Can switch any visual angle according to user's request, and output on the display screen.Import real time data into the three-dimensional virtual scene display module, intuitively 3-D display table tennis position and game status.Can access historical record, the historical flight path of simulation table tennis is to reach the replay effect.Can access associated storage information such as historical drop point and rotation, and switch the visual angle arbitrarily, output in the display screen according to user's needs.

The present invention is directed to trainer (referee in the match), set up user interactive module, set, statistics and demonstration at the quantity of information of being concerned about.

Ping-pong contest vision provided by the invention supports and match analytic system and system operation method that described operating procedure is as follows:

(1) builds hardware system, each camera interior and exterior parameter of semi-automatic demarcation.

(2) study table tennis model parameter, and online correction.

(3) filtering of vision collecting table tennis data and match.

(4) filtered online data is separated track, statistics one way track critical information, undated parameter.

(5) according to user interactions instruction statistical dependence information, revise or stored data.

(6) result data in real time input three-dimensional virtual scene is reappeared module, real-time play or specify playback in external display screen by interaction page; And while display process result in interaction page.

Wherein add up one way track critical information in the said method, undated parameter adopts following method:

(1) force analysis by table tennis obtains kinematics model, generates state-transition matrix and observing matrix, obtains its equation of motion, is used for online in real time filtering and match semireal time.

(2) set up learning machine, the parameter blackbox model, each parameter model of state-transition matrix and observing matrix obtains.

(3) fitting data obtains one way track critical information.

(4) data after the match are searched for intercepting, by the anti-parameter that pushes away of continuous model, and and discrete model correlation filtering improve parameter unification.

(5) parameter model being carried out online study upgrades.

Wherein one way track critical information is obtained the following method that adopts in the said method:

1. the filtering speed when the velocity information unification is passed through batting side's quarter by this track is the benchmark storage;

2. first the effective drop point the when unification of drop point information arrives the batting opponents' half by this track is the benchmark storage;

3. calculate rotation information based on rotating model, the record rotating vector;

4. judge serving side and inefficacy ball, simplify judge's operation.

Native system is owing to adopt image acquisition and the table tennis order claims to discern locating module and produces original image in real time and read in noisy three-dimensional data; Again the data of new generation inputs KALMAN filter module, export filtered afterwards the three-dimensional position result and initial speed and collision information; Data after the filtering enter track data match device module and carry out the adjustment of data, make data become more level and smooth, be close to actual value; Characteristics analysis module obtains the data after the match, carries out information such as length, sense of rotation, high low speed, falling sphere position, first end speed, flight time; The learning parameter module is adjusted the parameter of KALMAN filter module by the above falling sphere falling sphere position that obtains, first end speed, these three amounts of flight time; After the speed of receiving the output of KALMAN filter module in real time and locality, judge that by the one way judge module whether one way finishes; Filtering data is directly imported three-dimensional virtual scene reproduction module and can be reappeared on the reality screen; Other statistics is then stored by match characteristic memory module, by with the message transmission of user interactive module, finish the modification of data.

System of the present invention can be in the training and competition of table tennis, real-time processing data and in addition filtering and record; Have very friendly human-computer interaction interface, add up and analyze the victory or defeat situation of whole heat both sides batting, finish table tennis course of action Feature Extraction automatically, and can conveniently judge and carry out playback, be applied to ping-pong contest judge and television relay automatically; Have perfect, effective, comprehensive post-game analysis function, can extract and store characteristic information during the games automatically; The course of action feature that must add up table tennis player's service and win the match directly perceived, auxiliary trainer summarizes to a sportsman's batting custom and weakness, and later training is had reference value, can be applicable in the training of paddler's daily workout and pre-games.

The characteristics of native system maximum are by real-time vision collecting, develop a cover and have friendly and perfect information extraction storage and user interactions support.Can be good at carrying out track record (comprise filtering after position, velocity information), sportsman's characteristic information (comprising the point of impingement, rotation information, velocity reversal, velocity magnitude information, service method, the track etc. of winning the match) of batting extracts.

The present invention has very real use value:

Image data is applied to paddler's training in real time, analyzes and storing moving person play ball characteristics and inferior position, and must assemble for training for trainer's specific aim provides information, and good reference role is arranged.And can in ping-pong contest, judge automatically and reset, be applied to television relay.

Therefore, the present invention is a kind of very practical, effective high speed table tennis training back-up system, has good application prospects.

Description of drawings

Fig. 1 system architecture diagram

Fig. 2 system architecture process flow diagram

Fig. 3 program run surface chart

Fig. 4 table tennis force analysis figure, Fig. 9 A are the perspective view of table tennis track on the X-Z plane, and Fig. 9 B is table tennis process force analysis figure

Fig. 5 table tennis track filter effect figure, Figure 10 A are observed readings before the filtering, and Figure 10 B is the filtering result

Fig. 6 three-dimensional virtual scene emulation module work synoptic diagram

Fig. 7 screw is judged and information extraction

Fig. 8 filtering parameter obtains

Fig. 9 filtering parameter upgrades

Figure 10 container storage structure

Embodiment

Below in conjunction with accompanying drawing, describe the present invention in detail and utilize the twin camera real-time positioning, realize that the track of table tennis generates and the course of action analysis; The Real-Time Filtering and the record of movement locus three-dimensional information amount; Add up and analyze the victory or defeat situation of whole heat both sides batting; The automatic table tennis of must finishing is tentatively judged, and is reappeared in the scene of virtual three dimensional space; Mutual at basic enterprising pedestrian's machine of automatically judging, be can the reserve judge convenient and swift must judge and record, trainer to the embodiment of the method and system of each sportsman's training analysis.

Fig. 1 has provided the structured flowchart of system, total system mainly by: two above video cameras 1, control multiple-cameras are adopted the synchronous device 2 of figure synchronously, and image data acquiring transmitting device 3, the main control computer 4 and the externally connected with display screen 6 that include the system environments 5 of vision running software constitute.The system environments 5 of described vision running software comprises acquisition software, video image process software, polyphaser information fusion software, filtering software of the driving of video camera under different operating system, camera video etc., and all are hidden in Core Feature software intersection under the vision software users operation interface, and it is the bridge that hardware system exchanges with the user.The system environments of vision running software and hardware system have been formed each functional module of whole vision system.Video camera 1 is fixed on the ping-pong table sky, makes visual field intersection cover the effective coverage of table tennis; Synchronous device 2 produces the multichannel periodic pulse signal, connects video camera 1 by control line, and the control multiple-camera is adopted figure synchronously; Video camera 1 links to each other with the image data acquiring card by data line; The pci bus of data collecting card by main control computer is sent to processor with the data that collect and handles; Result outputs on the externally connected with display screen of main control computer and shows.Wherein, image acquisition is with take pictures image under adopting of the speed of 60hz by camera, through table tennis Target Recognition locating module, promptly be identified in the result of image by the HLS characteristic quantity of table tennis under stabilized light source, the confidential reference items by camera carry out three-dimensional and reduce and position.

Fig. 2 is the framework process flow diagram of system.Described system comprises following composition module: image acquisition and table tennis Target Recognition locating module, KALMAN filter module, track data match device module, characteristics analysis module, kinematic parameter learning machine module, one way judge module, three-dimensional virtual scene reappear module, match characteristic memory module; User interactive module;

The framework flow process of system is: the data that the pci bus of data collecting card by main control computer collects video camera 1 are sent to main control computer 4, through image acquisition and table tennis Target Recognition locating module Target Recognition, space orientation.

Then the data of real-time generation are sent into the KALMAN filter module and carry out Real-Time Filtering,, make data reliable more and level and smooth through the noise that the data after the filtering can effectively must reduce to introduce in the vision collecting.Characteristics analysis module is analyzed the movement locus and the current motion state (position, speed and collision) of table tennis.

The one way judge module passes through one way and judges (being whether table tennis is impacted by the other side), under the situation that one way continues, sends into three-dimensional virtual scene reproduction module in real time and carries out three-dimensional virtual scene simulation reproduction; Under the situation that one way finishes, track data match device module is carried out match with KALMAN match device, is position and speed match based on KALMAN filter state equation, makes data more level and smooth, and the noise that introduce filtering identification and location is near time of day.

The benefit of match track is the further data that must make to become reliable and level and smooth on original filtering basis.Filtering accuracy depends on the degree of agreeing with of model parameter.And in practice, this parameter is unknown often, need obtain by machine learning.Through data fitting, needing in the track can to obtain each characteristic quantity of handling.The track extracting parameter is sent into kinematic parameter learning machine module learn, can in operational process, just carry out effectively the deviation of whole model generation data.

The message instruction that user interactive module is imported into by the user can be interfered the storage and the three-dimensional simulation interface of data in real time.

Match characteristic memory module data structure storage module is the data structure storage module based on STL, the message instruction control of being imported into by the user and position, the velocity information of storing moving track characteristic; The sportsman who extracts in every movement locus the bat point of impingement, rotation, velocity reversal, size, the service method of feature and the track relevant information of winning the match.

Fig. 2 is the program run surface chart.Here be the interface of system and user interactions.The interface can be divided into 7 functional areas substantially.Zone " referee is mutual " is that the function that system user is controlled the match data structure by software realizes.Here controlled importing into of vision data.Zone " scoreboard " is the interface that intuitively shows ping-pong contest local exchange and historical game; Judge that by score control carries out ace-shot control here simultaneously.Zone " drop point demonstrations " is for the consideration to the actual demand of competing, and has strengthened the statistics Presentation Function to drop point.Drop point by the scoreboard left side shows controller chassis, can pointwise or disposable drop point and the distribution that shows statistics in the 3-D display window.And at this for observation history course of action track more directly perceived, software has increased the playback function of table tennis storage track and winning ball track, the course of action track of simulation playback appointment in the 3-D display window.Wherein, " match initial setting up ", " off-line reads control ", " statistical information setting ", " historical record " are as Accessibility Options.Zone " match initial setting up " is the initialization setting that a match is carried out.Here can set pattern, match rule and the entrant's of match situation.Zone " off-line reads control " is the instrument that the match or the training data of original storage are operated, and has realized breaking away from the off-line analysis and the simulation playback of vision system.Can read in by file and simulate playback and analysis; And can be by writing as file storage current match or training original data storage are got off.Zone " statistical information setting " is carried out the statistic analysis to sportsman's course of action provides interactive interface.Here the user can be respectively carries out the statistical study of drop point, rotation, speed and direction to the sportsman both sides of match (training).And show the designated statistics amount in " statistical information demonstration " zone, concrete distribution situation is come out with the 2 d plane picture match to the scalar statistics; Adopt 3 dimensional drawing intuitively to show to vector.Zone " historical record " is the real-time demonstration to current operation and vision input record.By interactive interface, system can effective, comprehensive data analysis that must play.Coach's analysis and sportsman's improvement all there is very strong realistic meaning.The statistics playback of the statistics of drop point and winning ball can be effective, convenient and swift batting custom and the weakness that obtains a sportsman.

Fig. 3 is table tennis force analysis figure.Wherein Fig. 3 A is the perspective view of table tennis track in the direction of motion vertical plane, and Fig. 3 B is table tennis process force analysis figure.Here, supposition table tennis earlier without spin.Concrete pivot analysis can specify in the back.The independent usefulness on 3 directions of here moving, and windage power and table tennis speed are proportional, and below for simplicity with x, the z axle is that example describes.The y axle is identical with the x axle.

Force analysis:

F _x＝f _x＝-kV _x

F _z＝(f _b-mg)+f _z＝(f _b-mg)-kV _z

Acceleration:

$a_x=\frac{F_x}{m}=-\frac{k}{m}{V}_x$

$a_z=\frac{F_z}{m}=\frac{-k}{m}{V}_z+(\frac{f_b}{m}-g)$

Remember kinematical equation

$V_x\left(t\right)=V_{x0}+{\∫}_0^t{a}_x\left(\mathrm{\τ}\right)\mathrm{d\τ}=V_{x0}-{\∫}_0^t\frac{k}{m}{V}_x\left(\mathrm{\τ}\right)\mathrm{d\τ}$

$V_z\left(t\right)=V_{z0}+{\∫}_0^t{a}_z\left(\mathrm{\τ}\right)\mathrm{d\τ}=V_{z0}+{\∫}_0^t(\frac{f_b}{m}-g-\frac{k}{m}{V}_z\left(\mathrm{\τ}\right))\mathrm{d\τ}=V_{z0}+(\frac{f_b}{m}-g)t-{\∫}_0^t{V}_z\left(\mathrm{\τ}\right)\mathrm{d\τ}$

Order $k=\frac{k}{m}$

Finally obtain:

$\left\{\begin{array}{c}{V}_x\left(t\right)=V_{x0}{e}^{-\mathrm{kt}}\\ V_y\left(t\right)=V_{y0}{e}^{-\mathrm{kt}}\\ V_z\left(t\right)=[V_{z0}+(\frac{f_b}{\mathrm{mk}}-\frac{g}{k}){]e}^{-\mathrm{kt}}-(\frac{f_b}{\mathrm{mk}}-\frac{g}{k})\end{array}\right.$

$\left\{\begin{array}{c}{S}_x\left(t\right)={\∫}_0^t{V}_x\left(\mathrm{\τ}\right)\mathrm{d\τ}=\frac{V_{x0}}{k}(1-e^{-\mathrm{kt}})\\ S_y\left(t\right)={\∫}_0^t{V}_y\left(\mathrm{\τ}\right)\mathrm{d\τ}=\frac{V_{y0}}{k}(1-e^{-\mathrm{kt}})\\ S_z\left(t\right)={\∫}_0^t{V}_z\left(\mathrm{\τ}\right)\mathrm{d\τ}=\frac{V_{z0}+(\frac{f_b}{\mathrm{mk}}-\frac{g}{k})}{k}(1-e^{-\mathrm{kt}})-(\frac{f_b}{\mathrm{mk}}-\frac{g}{k})t\end{array}\right.$

Be the collision model in the continuous model below:

Obtain collision time t by following formula through iteration _{Col_z}

$\left\{\begin{array}{c}{{\mathrm{vx}}_{\mathrm{colli}}}^{-}=V_{x0}{e}^{-k{t}_{\mathrm{col}\_z}}\\ {{\mathrm{vy}}_{\mathrm{colli}}}^{-}=V_{y0}{e}^{-{\mathrm{kt}}_{\mathrm{col}\_z}}\\ {{\mathrm{vz}}_{\mathrm{colli}}}^{-}=[V_{z0}+(\frac{f_b}{\mathrm{mk}}-\frac{g}{k})e^{-{\mathrm{kt}}_{\mathrm{col}\_z}}-(\frac{f_b}{\mathrm{mk}}-\frac{g}{k})\end{array}\right.$

$\left\{\begin{array}{c}{{\mathrm{vx}}_{\mathrm{colli}}}^{+}={{\mathrm{vx}}_{\mathrm{colli}}}^{-}\×\mathrm{colliconfX}\\ {{\mathrm{vx}}_{\mathrm{colli}}}^{+}={{\mathrm{vy}}_{\mathrm{colli}}}^{-}\×\mathrm{colliconfY}\\ {{\mathrm{vz}}_{\mathrm{colli}}}^{+}={{\mathrm{vz}}_{\mathrm{colli}}}^{-}\×\mathrm{colliconfZ}\end{array}\right.$

The non-stop flight model can get through discretize:

X _i+1＝AX _i+BU _i+w _i

Y _i＝HX _i+v _i

Wherein, quantity of state comprises position p and speed v, systematic error w, and observational error is v.

Here, state equation parameter and kinematical equation parameter can transform mutually, keep model unified.

Fig. 4 is table tennis track filter effect figure, and Fig. 4 A is an observed reading before the filtering, and Fig. 4 B is the filtering result.The discretization model of kinetics equation has shown that system state amount (position, speed) is with adopting the process that the figure cycle passes.With the KALMAN wave filter table tennis state is carried out Real-Time Filtering.As can be seen, there is certain observational error in observed reading before the filtering, filtered table tennis track, through the correction that the KALMAN wave filter of setting up based on top model equation carries out the table tennis process effectively, can effectively must make data become reliable and level and smooth.

Fig. 5 is a three-dimensional virtual scene emulation module work synoptic diagram.This module can be in real time and off-line show the position and the drop point analysis result of table tennis.At the camera collection image and after carrying out Target Recognition and space orientation, just can be sent in the three-dimensional artificial virtual scene module by the udp agreement.In the space, table tennis court of simulation, the current position display of table tennis is come out.Can reconstruct ping-pong contest three-dimensional scenic, comprise room, floor, ball table etc., and can specify to show the table tennis position, can realize regulating according to first input command of user and watch visual angle and far and near, automatic corner and course of action plane to follow the tracks of, function such as full frame.For the user's data analysis provides very friendly display interface.Fig. 5 A is the design sketch of one section real-time follow-up table tennis track; Fig. 5 B is the drop point distribution signal of game.All pick up from the design sketch of virtual scene.

Fig. 6 is that screw is judged and information extraction.In the component of 6D, shown the force analysis of table tennis rotation at Fig. 6 A.The object that rotates at fluid (in this example for air), with bear that fluid gives with the perpendicular power of direction of motion.Be magnus force, this power makes ball produce drift.If making a concerted effort not by the centre of sphere of batting, then ball is producing the proal while, also produces to rotatablely move.This is very common in the table tennis Chinese style.Analysis to rotation is most important in training.Therefore, accomplish qualitative and quantitative analysis to rotation information.Fig. 6 A is for to be respectively top spin, chop, hook and Slice to 6D.Can see the path of track and the difference without spin the time.So just can judge screw information in the prediction at track initial stage and the position deviation vector size after the actual acquisition track, and then obtain the rotation vector of every orbiting motion of table tennis by the equation of motion of continuous model.

Fig. 7 is the schematic three dimensional views that filtering parameter obtains.At first, track is carried out the KALMAN match.Here can pass through track fitting, reduce noise, the especially error of velocity information amount of parameter raw data; Parameter acquiring method:, intercept the table tennis state of flight and the collision state of flight (colliding the last cycle) constantly of the predicted data transmitting time point in the track according to the discrete model of table tennis collisionless flight.Obtain this two states and periodicity, be converted into continuous model, use following formula:

$\left\{\begin{array}{c}{S}_{\mathrm{xy}}\left(t\right)={\∫}_0^t{V}_{\mathrm{xy}}\left(\mathrm{\τ}\right)\mathrm{d\τ}=\frac{V_{\mathrm{xy}0}}{k}(1-e^{-\mathrm{kt}})\\ S_z\left(t\right)={\∫}_0^t{V}_z\left(\mathrm{\τ}\right)\mathrm{d\τ}=\frac{V_{z0}+(\frac{f_b}{\mathrm{mk}}-\frac{g}{k})}{k}(1-e^{-\mathrm{kt}})-(\frac{f_b}{\mathrm{mk}}-\frac{g}{k})t\end{array}\right.$

The result of inverse tries to achieve parameter g and k, and the unified conversion relation by discrete model and continuous model, but just exposure parameter analytic solution.

The blackbox model that Fig. 8 upgrades for filtering parameter.Set up a kinematic parameter learning machine here based on rbf (radial basis function) network; data analysis is as input to the tangential velocity of original state (future position) and the elevation angle; the parameter that every track obtains as output as training sample; set up a blackbox model, obtained parameter model as far as possible accurately by study.Here model uses the basic algorithm of the learning-oriented radial primary function network of LR-OLS (Local regularisation) as parameter learning, and strengthening the robustness of network in conjunction with the D-Optimality algorithm, foundation can be efficiently and the accurate machine learning mode of learning parameter model; Related modeling is carried out in coupling to Correlative Influence Factors; By the model error of suitable sacrifice sample training, make the generalization ability of model reach the highest; And in the process of table tennis track data accumulation, bring in constant renewal in and learn.

Figure 9 shows that the container storage structure.The emphasis that competition data is handled is must handle data at all levels the time, can pass through user interactive module effective, convenient and swift carries out calling and revising of data; Difficult point is to adopt in real time in figure and the location process at video camera, can modularization gets correctly to handle vision input data and the user sends message, and data are carried out a series of judgement and analysis.Therefore, here inside the class of Feng Zhuan called after GameAnalyze, this class has mainly encapsulated in the match for the data of real-time input, the function that messages sent by users is handled, and defined the storage organization of all quantity of information of competing, consider from data structure, be with four containers: the in real time current one way of input, current match point, current office and current match main composition, and encapsulated the related function of data processing in four containers respectively.Can activate correlative code in the time of user's transmission " score ", " nothing ball ", " local exchange end " and " end of match " message, the effective information of the data of bottom container and this data extract is changed in high one deck container.Whereby the quantity of information of whole schedules in the ping-pong contest there has been a management comprehensively and effectively, judicatory in the ping-pong contest has been wanted the track datas such as drop point site, pivot analysis result of playback have one reasonably to store.And " drop point demonstration " zone in user interactive module, point to current data module by controlling a pointer, but pointwise or disposablely show history realizes the reproduction of drop point and track in the 3-D display window.

Claims (6)

1. a ping-pong contest vision is supported and the match analytic system, described system is mainly adopted the synchronous device of figure synchronously by two above video cameras, control multiple-camera, image data acquiring transmitting device, the main control computer that includes the system environments of vision running software, externally connected with display screen are formed; Video camera is fixed on the ping-pong table sky, and visual field intersection covers the effective coverage of table tennis, carries out image acquisition, and synchronous device produces the multichannel periodic pulse signal, connects video camera by control line, and the control multiple-camera is adopted figure synchronously; Video camera links to each other with the image data acquiring card by data line; Data collecting card is by the bus of main control computer, and the processor that the data that collect is sent to the image data acquiring transmitting device is handled; Result outputs on the externally connected with display screen of main control computer and shows; It is characterized in that system comprises following composition module: image acquisition and table tennis Target Recognition locating module, KALMAN filter module, track data match device module, characteristics analysis module, kinematic parameter learning machine module, one way judge module, three-dimensional virtual scene reappear module, match characteristic memory module; User interactive module; Described system architecture flow process is: the main control computer epigraph is gathered and table tennis Target Recognition locating module, through Target Recognition, space orientation, the data that generate is in real time sent into the KALMAN filter module carry out filtering; The one way judge module is judged, under the situation that one way continues, sends into three-dimensional virtual scene in real time and reappears module simulation reproduction three-dimensional virtual scene; Under the situation that one way finishes, track data match device module is carried out match with KALMAN match device, KALMAN match device carries out match and generates the movement locus feature, characteristics analysis module is analyzed the movement locus and the current motion state information of table tennis, characteristics analysis module obtains the data after the match, send into kinematic parameter learning machine module, described kinematic parameter learning machine module is adjusted the parameter of KALMAN filter module, and the movement locus feature of generation is sent into record match characteristic memory module record; The message instruction that user interactive module is imported into by the user is interfered the storage and the three-dimensional simulation interface of data in real time.

2. ping-pong contest vision according to claim 1 is supported and the match analytic system, it is characterized in that described match characteristic memory module is the match characteristic memory module based on STL, position, the velocity information of control and storing moving track characteristic; The sportsman who extracts in every track the bat point of impingement, rotation, velocity reversal, size, the serving side of feature and the relevant information of the track of winning the match.

3. ping-pong contest vision according to claim 2 is supported and the match analytic system, it is characterized in that described match characteristic memory module mainly is that the local exchange information data structure of the ace-shot information data structure with the trace information data structure of current one way, current match point, current office and four containers of information for the game data structure of current match constitute, and encapsulated the related function of data processing in four containers respectively; Activate correlative code when sending instruction message by the user, the effective information of the data of bottom container and this data extract is changed in high one deck container, to the quantity of information management of whole schedules in the ping-pong contest, simultaneously judicatory in the ping-pong contest is wanted drop point site, the pivot analysis results trace data storage of playback; Current data module is pointed to by steering needle in drop point viewing area in user interactive module, and pointwise or disposablely show history in the 3-D display window realizes the reproduction of drop point and track.

4. a ping-pong contest vision according to claim 1 is supported and the operation method of match analytic system, and step is as follows:

(1), build two video cameras, the image data acquiring transmitting device, the system equipment of main control computer is demarcated each camera interior and exterior parameter;

(2), off-line learning and definite table tennis model parameter, and online correction;

(3), system operation, KALMAN wave filter, the track data match device of gathering the table tennis data carry out filtering and match;

(4), filtered online data is separated track statistics one way track critical information, undated parameter;

(5), according to user interactions instruction statistical dependence information, modification or stored data;

(6) result data in real time input three-dimensional virtual scene is reappeared module, real-time play or specify playback in external display screen by interaction page; And while display process result in interaction page.

5. want the operation method of 4 described ping-pong contest visions supports and match analytic system according to right, it is characterized in that described statistics one way track critical information, the undated parameter method is as follows:

A) force analysis by table tennis obtains kinematics model, generates state-transition matrix and observing matrix, obtains its equation of motion, is used for online in real time filtering and match semireal time;

B) set up learning machine, the parameter blackbox model, each parameter model of state-transition matrix and observing matrix obtains;

C) fitting data obtains one way track critical information;

D) data after the match are searched for intercepting, by the anti-parameter that pushes away of continuous model, and and discrete model correlation filtering improve parameter unification;

E) parameter model being carried out online study upgrades.

6. want the operation method of 4 described ping-pong contest visions supports and match analytic system according to right, it is characterized in that the described one way track critical information method of obtaining:

Filtering speed when 1) the velocity information unification is by real-time input trajectory process batting side quarter is the benchmark storage;

First effective drop point when 2) unification of drop point information arrives the batting opponents' half by real-time input trajectory is the benchmark storage;

3) calculate rotation information based on rotating model, the record rotating vector;

4) judge serving side and inefficacy ball.

Images