JP2008022142A - Method, apparatus and program for detecting pitching event, and computer-readable recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To dispense with division of a video image into segments of one pitcher in detecting a pitching event of pitchers from the video image of a baseball game. <P>SOLUTION: The pitching event detection method acquires pitching event candidates including time information, extracts an operation characteristic of the video image, uses the time information of the pitching event candidates to select operations of the pitching event candidates on the basis of the operation characteristics, establishes a limited segment width and how to move the limited segment width, extracts pitching operation characteristics within all the established limited segment widths, determines the pitching operation characteristics utilized by a media time on the basis of the pitching operation characteristics, uses the pitching operation characteristics and the operation characteristics to discriminate whether or not a motion is a pitching operation, and outputs an event of the pitching operation when the motion is the pitching operation. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a pitching event detection method and apparatus, a program, and a computer-readable recording medium, and more particularly, a pitching part (from a baseball video) using a multimodal processing technology of sound and images in video content processing technology. The present invention relates to a pitching event detection method and apparatus for detecting a pitching event), a program, and a computer-readable recording medium.

  As a first conventional technique for detecting a pitch from a baseball video, a technique for identifying a pitch event by detecting a green area or a brown area in the video and detecting a composition when the pitcher throws (For example, refer to Patent Document 1).

In addition, as a second conventional technique, a technique for detecting a pitcher's pitching portion with higher accuracy than unimodal analysis by using acoustic analysis and image analysis cooperatively instead of composition has been proposed. (For example, refer nonpatent literature 1).
JP 2003-52003 A Amami Mikami, Seiichi Sugaya, Masashi Morimoto “Throwing event detection from baseball video using sudden sound detection and unsupervised clustering” PRMU2004-139, pp.31-36

  However, in the first conventional method, it is necessary to create an exhaustive template in advance according to the ground situation and the lighting situation (digame, night game) of the game.

  On the other hand, the second conventional method can automatically create a pitching action template according to the video, but it cannot respond to the change of the pitcher, so the baseball video must be input after dividing it into one pitcher section. There is a disadvantage that it must be.

  The present invention has been made in view of the above points. In the detection of a pitcher's pitching portion (hereinafter referred to as a pitching event) from a baseball video, the pitching operation characteristics are dynamically obtained by performing automatic temporal division. It is an object of the present invention to provide a pitching event detection method and apparatus, a program, and a computer-readable recording medium that can acquire a baseball video without being divided for each pitcher.

  FIG. 1 is a diagram for explaining the principle of the present invention.

The present invention (Claim 1) is a pitching event detection method for detecting a pitcher's pitching portion (throwing event) from a baseball video,
A pitching event candidate acquisition unit acquires a pitching event candidate including time information and stores it in a pitching candidate event storage unit (step 1);
A motion feature extraction step (step 2) in which the motion feature extraction means extracts the motion feature of the video;
The pitching event candidate motion feature selection means uses the pitch event candidate time information stored in the pitching candidate event storage means and the motion feature extracted in the motion feature extraction step (step 2), and A pitching event candidate motion feature selection step (step 3) for selecting a pitching event candidate motion in the storage means;
A limited section setting means for determining a limited section width and how to move the limited section width (step 4);
The in-sector pitching motion feature acquisition means extracts the pitching motion features in all the limited zones set in the limited zone setting step from the pitch event candidate motions selected in the pitch event candidate motion feature selection step. An intra-section pitching motion feature acquisition step (step 5) stored in the motion feature storage means;
A pitching motion feature selection step (step 6) in which the pitching motion feature selection means determines the pitching motion feature to be used by the media time T from the pitching motion feature stored in the pitching motion feature storage means
The determination means uses the pitching motion feature selected in the pitching motion feature selection step (step 6) to determine whether or not the motion feature extracted in the motion feature extraction step (step 2) at the media time T is a pitching motion. If it is a pitching operation, a determination step (step 7) for outputting an event of the pitching operation is performed.

Further, the present invention (Claim 2) provides a pitch event candidate acquisition step (Step 1).
Sudden pronunciation is used when acquiring pitching event candidates.

Further, according to the present invention (Claim 3), in the motion feature extraction step (Step 2),
Motion History Image is used as an operation feature.

In the present invention (Claim 4), in the in-section pitching motion feature acquisition step (Step 5),
The motion history image of the motion feature in the motion feature extraction step (step 2) is subjected to reduction / enlargement / labeling processing to create a motion history image for each object.

In the present invention (Claim 5), in the pitching motion feature selection step (Step 6),
It uses not only the current media time but also the throwing motion characteristics in the media time in the vicinity.

Further, the present invention (Claim 6) is a limited section setting step (Step 4),
The number of pitching event candidates is used when setting a limited section.

Further, according to the present invention (Claim 7), in the limited section setting step (Step 4),
A fixed media time interval is used when setting a limited section.

Further, the present invention (Claim 8), in the limited section setting step (Step 4),
When moving the limited section, shift by a certain number of pitching event candidates.

Further, the present invention (Claim 9), in the limited section setting step (Step 4),
When moving the limited section, shift by a certain media time.

  FIG. 2 is a principle configuration diagram of the present invention.

The present invention (Claim 10) is a pitching event detection device for detecting a pitching part (throwing event) from a baseball video,
Video storage means 12 for storing video;
A pitching event candidate acquisition unit 1 that acquires a pitching event candidate including time information and stores the pitching event candidate in the pitching candidate event storage unit 8;
Motion feature extraction means 2 for extracting motion characteristics of video in the video storage means 12;
Using the time information of the pitching event candidate stored in the pitching candidate event storage unit 8 and the motion feature extracted by the motion feature extraction unit 2, the operation of the pitching event candidate in the pitching candidate event storage unit 8 is selected. A pitching event candidate motion feature selection means 11 to perform,
Limited section setting means 3 for determining the limited section width and how to move the limited section width;
The pitching motion features in all the limited zones set by the limited zone setting unit 3 are extracted from the pitching event candidate motions selected by the pitching event candidate motion feature selection unit 11 and stored in the pitching motion feature storage unit 9. In-section pitching motion feature acquisition means 4;
A pitching motion feature selection unit 6 for determining a pitching motion feature to be used according to the media time T from the pitching motion feature stored in the pitching motion feature storage unit 9;
Using the pitching motion feature selected by the pitching motion feature selection means 6, it is determined whether or not the motion feature extracted by the motion feature extraction means 2 at the media time T is a pitching motion. Has a judging means 7 for outputting an event of the pitching operation.

  Further, according to the present invention (claim 11), the pitch event candidate acquisition unit 1 uses sudden sound when acquiring a pitch event candidate.

  In the present invention (Claim 12), the motion feature extraction means 2 uses Motion History Image as motion features.

  Further, according to the present invention (claim 13), the intra-section pitching motion feature acquisition unit 4 performs reduction / enlargement / labeling processing on the motion history image of the motion feature of the motion feature extraction unit 2 to perform motion for each object. Create a History Image.

  In the present invention (Claim 14), the pitching motion feature selection means 6 uses not only the current media time but also the pitching motion feature in the media time in the vicinity thereof.

  In the present invention (Claim 15), the limited section setting means 3 uses the number of pitching event candidates when setting the limited section.

  In the present invention (Claim 16), the limited section setting means 3 uses a fixed media time interval when setting the limited section.

  According to the present invention (Claim 17), when the limited section is moved in the limited section setting means 3, it is shifted by a certain number of pitching event candidates.

  In the present invention (Claim 18), when the limited section is moved in the limited section setting means 3, it is shifted by a certain media time.

  The present invention (Claim 19) is a pitching event detection program for causing a computer to execute each means of the pitching event detection device according to Claims 10 to 18.

  The present invention (Claim 20) is a computer-readable recording medium storing a pitching event detection program for causing a computer to execute each means of the pitching event detection device according to Claims 10 to 18.

  According to the first conventional method, when a pitching event is detected from a baseball video, it is necessary to input the baseball video after dividing it into one pitcher section, which is inconvenient as baseball video indexing. On the other hand, according to the present invention, it is not necessary to divide the video into sections of one pitcher, and baseball video indexing using a pitching event can be performed with high accuracy and less time for input.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  FIG. 3 shows a configuration of a pitching event detection apparatus according to an embodiment of the present invention.

  The apparatus shown in the figure includes a pitching event candidate acquisition unit 1, a motion feature extraction unit 2, a limited section setting unit 3, an intra-section pitching motion feature acquisition unit 4, a pitching motion feature selection unit 6, a determination unit 7, and a pitching event candidate storage. Unit 8, a pitching event storage unit 10, a pitching event candidate motion feature selection unit 11, and a video storage unit 12.

  The pitching event candidate storage unit 8 stores the pitching event candidate and time information acquired by the pitching event candidate acquisition unit 1.

  The pitching motion feature storage unit 9 stores the pitching motion feature acquired by the in-section pitching motion feature acquisition unit 4.

  The pitching event storage unit 10 stores a pitching event that is a result of the present apparatus.

  The video storage unit 12 stores video.

  The pitching event candidate acquisition unit 1 obtains a pitching event candidate by some method and stores it in the pitching event candidate storage unit 8.

  The motion feature extraction unit 2 extracts motion features from the video in the video storage unit 12.

  The pitch event candidate motion feature selection unit 11 selects the pitch event candidate motion extracted by the motion feature extraction unit 2 using the pitch event candidate time information stored in the pitch event candidate storage unit 8, and stores the memory Temporarily stored in a storage means such as

  The limited section setting unit 3 determines the limited section width and how to move the limited section width.

  The intra-section pitching motion feature acquisition unit 4 extracts the pitching event candidate motions extracted by the pitching event candidate motion feature selection unit 11 and stored in the memory or the like in all the limited sections set by the limited section setting unit 3. The pitching motion feature is extracted from the pitching motion feature and stored in the pitching motion feature storage unit 9.

  The pitching motion feature selection unit 6 determines a pitching motion feature to be used according to the media time from the pitching motion features stored in the pitching motion feature storage unit 9.

  The determination unit 7 determines whether or not the motion feature extracted by the motion feature extraction unit 2 is a pitching motion using the pitching motion feature selected by the pitching motion feature selection unit 6, and is determined to be a pitching motion. The pitching motion feature is stored in the pitching event storage unit 10.

  The operation in the above configuration will be described below.

  FIG. 4 is a flowchart of the pitching event detection operation according to the embodiment of the present invention.

  Step 101) The pitching event candidate acquisition unit 1 detects a pitching event candidate by any method. For example, it is conceivable to use sound generated by hitting or catching a video in the video storage unit 12. Further, the result is stored in the pitching event candidate storage unit 8. Note that the acoustic analysis uses non-patent literature 1 acoustic analysis to detect sudden sound, and the motion analysis uses the non-patent literature 1 MHI (Motion History Image). This holds a pitching event candidate ID and its time. In addition, reliability may be maintained.

  Step 102) The motion feature extraction unit 2 extracts video features (motion features) from the video storage unit 12. In this process, the method described in Non-Patent Document 1 can be used.

  Step 103) The pitch event candidate motion feature selection unit 11 selects a pitch event candidate from the motion features extracted in step 102 using the pitch event candidate time information stored in the pitch event candidate storage unit 8. .

  FIG. 5 is an example of the output of the pitching event candidate motion feature selection unit in one embodiment of the present invention. The example shown in the figure summarizes the output of the pitching event candidate motion feature selection unit 11 when using Non-Patent Document 1 described above, and stores the event candidate ID, the generated media time, and the MHI that is the motion feature. Etc. are stored in the storage means.

  Step 104) The limited section setting unit 3 sets a limited section. In the setting of the limited section, how to shift the limited section width and the limited section is set. Both the limited section width and the shifting method can be specified by the number or time of pitching event candidates. Either may be used, and designation by number and designation by time may be used together.

  Step 105) The intra-section pitching motion feature acquisition unit 4 acquires the intra-section pitching motion feature from the output of the pitching event candidate motion feature selection unit 11 for each limited section set in Step 104. Below, the operation | movement of the said pitching motion characteristic acquisition part 4 is demonstrated. FIG. 6 is a flowchart of the operation of the intra-section pitching motion feature acquisition unit according to the embodiment of the present invention.

  In step 105, the following processing from step 401 to step 405 is repeated for all limited sections defined in step 104.

  Step 401) The intra-section pitching motion feature acquisition unit 4 obtains the motion feature for each motion object existing in the frame from the output of the pitching event candidate motion feature selection unit 11 obtained in Step 103.

  Step 402) Cluster motion features for each object.

  Step 403) Obtain the maximum clustering result.

  Step 404) Determine whether the operating characteristics of the largest cluster are within a specified range. If it is within the range, the process proceeds to step 405. Otherwise, the process proceeds to step 401.

  Step 405) The motion feature of the maximum cluster is stored in the pitching motion feature storage unit 9 as the motion feature in the limited section.

  There is no particular limitation on the method for obtaining the motion characteristics for each motion object in step 401 here. For example, if MHI is used as a motion feature, it is possible to acquire MHI (object MHI) for each motion object by performing reduction / enlargement / labeling processing on the MHI. . A flowchart of reduction / enlargement / labeling is shown in FIG. First, the intra-section pitching motion feature acquisition unit 4 reduces the image to ¼ (step 701), and then enlarges the reduced image four times (step 702). Labeling using 8-connection is performed (step 703).

  In the clustering algorithm, any method may be used as long as the processing is terminated by the distance threshold. This is because the number of clusters (= the number of pitchers) is unknown and the number of clusters cannot be set as the clustering end condition.

  Through the above steps, the pitching motion characteristics that change according to the media time are stored in the pitching motion feature storage unit 9. FIG. 8 shows an example of the pitching motion feature that changes depending on the media time stored in the pitching motion feature storage unit 9. Hereinafter, this will be referred to as a dynamic pitching operation feature.

  As shown in FIG. 8, the dynamic pitching motion feature includes a pitching motion feature ID, a limited zone start time, a limited zone end time, and a pitching motion feature. That is, it is represented by a limited section and a set of pitching motion features acquired from the limited section. FIG. 9 shows a schematic diagram of FIG. 8 in the time axis direction.

  Step 106) Next, the pitching event not included in the pitching event candidates is also detected by re-searching the entire video in the video storage unit 12 using the dynamic pitching motion characteristics that change according to the media time. To do.

  In this step, according to the media time, the pitching motion feature used in the media time is selected from the dynamic pitching motion features and determined. The simplest information is a method using a pitching motion feature including the current media time or a pitching motion feature closest to the current media time.

  Further, another method for realizing the step 106 will be described with reference to FIG.

  FIG. 9 shows dynamic throwing motion characteristics in a certain video. In the figure, 1, 2 and 3 each indicate one pitching motion feature, and a bidirectional arrow below each pitching motion feature indicates a limited section used for calculation of each pitching motion feature. For example, in FIG. 9, it may seem intuitively that the pitching movement feature used at time A is only 1, but since there is a possibility that the pitcher change is always performed even in the limited section, With only the pitching motion feature 1, there is a possibility that the pitching event by the pitcher after the pitcher change may be missed. Taking this into consideration, FIG. 10 shows the pitching motion feature selection processing from the dynamic pitching motion feature in step 106.

  Step 1001) For all media times (time in the flowchart) T, the following processing from Step 1002 to Step 1012 is repeated.

  Step 1002) It is checked whether or not a pitching motion feature composed only of a media time earlier than the media time T exists in the dynamic pitching motion features. If it exists, the process proceeds to step 1003. Otherwise, the process proceeds to step 1004.

  Step 1003) Among the pitching motion features composed only of the media time earlier than the media time T, the pitching motion feature ID consisting of the latest media time is set to S.

  Step 1004) It is determined whether or not a pitching motion feature including the media time T exists. If it exists, the process proceeds to step 1006. Otherwise, the process proceeds to step 1006.

  Step 1005) Let S be the throwing motion feature ID including the media time T.

  Step 1006) S is set to 0.

  Step 1007) It is confirmed whether or not the pitching motion feature composed only of the media time later than the media time T exists in the dynamic pitching motion feature. If it exists, the process proceeds to step 1008. Otherwise, the process proceeds to step 1009.

  Step 1008) Let E be the pitching motion feature ID consisting of the earliest media time among the pitching motion features composed only of the media time later than the media time T.

  Step 1009) It is determined whether or not the pitching motion feature is present in the limited section including the media time T among the dynamic pitching motion features. If it exists, the process proceeds to step 610. Otherwise, the process proceeds to step 611.

  Step 1010) Let E be the throwing motion feature ID including the media time T.

  Step 1011) Let E be the throwing motion feature ID constituted by the latest media time.

  Step 1012) The pitching motion characteristics used at the media time T are set as pitching motions having IDs from S to E among the dynamic pitching motion characteristics.

  Step 107) The determination unit 7 performs a re-search on the entire video using the pitching motion feature selected and determined by the pitching motion feature selection unit 6 in Step 106.

  The operation of step 107 is shown in FIG.

  Step 1101) The determination unit 7 repeats Step 1102 to Step 1106 for all times.

  Step 1102) Step 1103 is performed for all the throwing motion characteristics determined in Step 106 when used at the media time T.

  Step 1103) The similarity between the motion feature at the media time T and the pitching motion feature is calculated.

  Step 1104) The similarity with the pitching motion feature showing the maximum similarity is set as the similarity at the media time T.

  Step 1105) It is determined whether the similarity at the media time T is greater than a threshold value. If larger, the process proceeds to step 1106.

  Step 1106) The fact that a pitching event is detected at the media time T is recorded in the pitching event storage unit 10.

  Embodiments of the present invention will be described below with reference to the drawings.

  In the present embodiment, the method of Non-Patent Document 1 based on sudden sound detection is used to obtain the pitch event candidate in Step 101 of FIG. 4 described above. In the technique of Non-Patent Document 1, sudden sound detection is performed by paying attention to the catching and hitting sound generated immediately after the pitcher's pitching.

  In step 102, Motion History Image is detected as an operation feature for the video. A motion history image (hereinafter abbreviated as MHI) is an image having motion history information. Pixels that have moved recently have brighter pixels and pixels that have moved older have darker pixel values, and are defined by the following equation. .

M (x, y, t) represents a pixel value at coordinates (x, y) at time t. Further, D (x, y, t) indicates the pixel value of the coordinates (x, y) of the inter-frame difference binarized image at times t and t−τ. F () is a function that returns the maximum argument. That is, the pixel value V is given to the pixel that has moved most recently, and the pixel value is decreased by 1 each time the unit time τ elapses. However, when a new movement occurs at the same coordinates, V is set because the if condition of the above equation is satisfied. As a result, an image having movement history information can be created.

  FIG. 12 shows an example of MHI. However, FIG. 12 is an image in which the brightness is reversed for easy viewing.

  In step 103 of FIG. 4, the pitching event candidate motion feature selection unit 11 selects a pitching event candidate portion from the motion features using the pitch event candidate time information. At this time, it may be necessary to specify an offset value from the pitch event candidate acquisition method and the motion feature acquisition method. In the case of the combination of sudden sound detection and MHI this time, an appropriate offset value is about -0.4 seconds.

  In step 104, the limited section setting unit 3 sets a limited section. In the present embodiment, as a method for setting the limited section width, a certain number of pitching event candidates is used. In addition, as a way of moving the limited section, it is assumed that a certain number of pitching event candidates are shifted in the time axis direction. However, as a method for setting the limited section width, a reference of a constant media time may be used, and as a method of movement, the movement may be performed in the time axis direction by a fixed media time.

  As a method for setting the limited section width, a fixed media time is basically used. However, if the number of pitching event candidates within the media time does not reach the predetermined number, the number of pitching event candidates becomes the preset number. A complex setting method such as extending the limited section width may be used.

  In step 105, the in-section pitching motion feature acquisition unit 4 acquires the pitching motion characteristics in each limited section set in step 104. In the present embodiment, first, the reduction / enlargement processing as shown in FIG. 7 is performed on the MHI that is the operation feature of each pitching event candidate, and labeling is performed. Obtain MHI (object MHI). An example of this is shown in FIG. FIG. 13A corresponds to an operation feature in one pitching event candidate. The finally obtained object MHI is (b) to (d).

  Next, clustering of the object MHI is performed. According to Non-Patent Document 1, a majority of pitching event candidates are correct pitching events. Therefore, after clustering, the maximum cluster is an object MHI that represents an operation characteristic related to a pitching operation. Also in this embodiment, the object MHI that has become the maximum cluster as a result of clustering is used as a pitching motion feature in each limited section.

  Many clustering algorithms have been proposed, but any method can be used. Here, the nearest neighbor algorithm and the single connection algorithm are used. However, the distance calculation between the objects MHI at the time of clustering was performed by the following equation.

Here, v ^a (x, y) represents the pixel value at the coordinates (x, y) of the object MHIa, and v ^b (x, y) represents the pixel value at the coordinates (x, y) of the object MHIb. In addition, X and Y select the smaller one of the width and height of the objects MHIa and MHIb.

  In the present embodiment, the distance between the objects MHI is calculated, but other distance definition methods such as a simple Euclidean distance may be used.

  In step 106 in FIG. 4, the pitching motion feature selection unit 6 determines from the dynamic pitching motion feature the pitching motion feature including the media time T that is the detection target and the nearest pitching motion feature including the media time T. Select a throwing motion feature. Thereby, even in a limited section including a pitcher change, it is possible to appropriately detect a pitching event using the pitching motion characteristics in the preceding and following sections.

  For example, it is possible to use all the pitching motion features of the dynamic pitching motion features at all times. In this case, although there are disadvantages in processing time and the like, it is considered that there is a great advantage in terms of the reproducibility of pitching event detection.

  Furthermore, a method of presenting all the pitching motion characteristics of the dynamic pitching motion characteristics to the user, prompting the user to determine whether to use the pitching motion characteristics, and using all the pitching motion characteristics determined to be used is also conceivable.

  Finally, determination is performed in the determination unit 7 in step 107. As a result of template matching, the portion where the distance from the template is equal to or less than the threshold value is stored in the pitching event storage unit 10.

  Through the above steps, a pitching motion feature can be dynamically extracted from a baseball video including a pitch change, and a pitch event can be detected with high accuracy.

  The present invention can construct the function of the pitching event detection device shown in FIG. 3 as a program and install it on a computer used as the pitching event detection device, or distribute it via a network. is there.

  Further, the constructed program can be stored in a portable storage medium such as a hard disk, a flexible disk, or a CD-ROM, and installed in a computer or distributed.

  The present invention is not limited to the above-described embodiments and examples, and various modifications and applications can be made within the scope of the claims.

  The present invention is applicable to a technique for detecting a pitching event from a baseball video, in particular, a multimodal processing technique.

It is a figure for demonstrating the principle of this invention. It is a principle block diagram of this invention. It is a block diagram of the pitching event detection apparatus in one embodiment of this invention. It is a flowchart of the pitching event detection operation | movement in one embodiment of this invention. It is an example of the output of the pitch event candidate operation | movement feature selection part in one embodiment of this invention. It is a flowchart of operation | movement of the pitching motion characteristic acquisition part in an area in one embodiment of this invention. It is a flowchart of the operation | movement which calculates | requires the operation | movement characteristic for every operation | movement object in one embodiment of this invention. It is an example of the pitching motion characteristic which changes with the media time stored in the pitching motion characteristic memory | storage part in one embodiment of this invention. It is the example which represented the dynamic pitching operation characteristic in one embodiment of this invention typically in the time-axis direction. It is a flowchart of the selection process of the pitching operation | movement characteristic in one embodiment of this invention. It is a flowchart of operation | movement of the determination part in one embodiment of this invention. It is an example of MHI of one Example of this invention. It is a figure which shows how to obtain | require MHI of one Example of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Throwing event candidate acquisition means, Throwing event acquisition part 2 Operation | movement feature extraction means, Operation | movement feature extraction part 3 Restriction area setting means, Restriction area setting part 4 Intra-zone pitching action characteristic acquisition means, Intra-zone pitching action characteristic acquisition part 6 Throwing action Feature selection unit, pitching motion feature selection unit 7 Determination unit, determination unit 8 Throwing event candidate storage unit, pitching event candidate acquisition unit 9 Throwing motion feature storage unit, pitching motion feature storage unit 10 Throwing event storage unit 11 Throwing event candidate motion feature Selection means, pitching event candidate motion feature selection section 12 Video storage means, video storage section

Claims (20)

A pitching event detection method for detecting a pitcher's pitching part (throwing event) from a baseball image,
A pitching event candidate acquisition unit acquires a pitching event candidate including time information and stores it in a pitching candidate event storage unit;
An operation feature extraction step in which an operation feature extraction means extracts an operation feature of the video;
The pitching event candidate motion feature selection means uses the pitching event candidate time information stored in the pitching candidate event storage means and the pitching candidate event storage means using the motion features extracted in the motion feature extraction step. A pitching event candidate motion feature selection step for selecting a pitching event candidate motion of
A limited section setting means for determining a limited section width and how to move the limited section width;
In-sector pitching motion feature acquisition means extracts pitching motion features in all the limited zones set in the limited zone setting step from the pitching event candidate motions selected in the pitching event candidate motion feature selection step. A pitching motion feature acquisition step for storing in the pitching motion feature storage means;
A pitching motion feature selection step for determining a pitching motion feature to be used by a media time T from the pitching motion feature stored in the pitching motion feature storage unit;
A determination unit determines whether the motion feature extracted in the motion feature extraction step at the media time T is a pitching motion using the pitching motion feature selected in the pitching motion feature selection step. If it is a pitching motion, a determination step for outputting an event of the pitching motion;
A pitching event detection method comprising: In the pitch event candidate acquisition step,
When acquiring the pitching event candidate, sudden sound is used,
The pitching event detection method according to claim 1. In the operation feature extraction step,
As the operation feature, Motion History Image is used.
The pitching event detection method according to claim 1 or 2. In the in-section pitching movement feature acquisition step,
The pitching event detection method according to claim 3, wherein the motion history image of the motion feature in the motion feature extraction step is subjected to reduction / enlargement / labeling processing to create a motion history image for each object. In the pitching operation feature selection step,
The pitching event detection method according to any one of claims 1 to 4, wherein not only the current media time but also a pitching operation characteristic at a media time in the vicinity thereof is added. In the limited section setting step,
When setting the limited section, the number of pitching event candidates is used.
The pitching event detection method according to claim 1. In the limited section setting step,
When setting the limited section, a fixed media time interval is used.
The pitching event detection method according to claim 1. In the limited section setting step,
When moving the limited section, shift by a certain number of pitching event candidates,
The pitching event detection method according to claim 1. In the limited section setting step,
When moving the limited section, shift by a certain media time,
The pitching event detection method according to claim 7. A pitching event detection device for detecting a pitcher's pitching part (throwing event) from a baseball image,
Video storage means for storing video;
A pitching event candidate acquisition unit that acquires a pitching event candidate including time information and stores the pitching event candidate in a pitching candidate event storage unit;
Motion feature extraction means for extracting motion characteristics of the video stored in the video storage means;
Using the time information of the pitching event candidate stored in the pitching candidate event storage unit and the motion feature extracted by the motion feature extraction unit, the operation of the pitching event candidate of the pitching candidate event storage unit is performed. A pitching event candidate motion feature selection means to select;
Limited section setting means for determining the limited section width and how to move the limited section width;
The pitching motion features in all the limited zones set by the limited zone setting means are extracted from the pitching event candidate motions selected by the pitching event candidate motion feature selection means and stored in the pitching motion feature storage means. In-section pitching motion feature acquisition means;
A pitching motion feature selection unit for determining a pitching motion feature to be used according to a media time T from the pitching motion feature stored in the pitching motion feature storage unit;
Using the pitching motion feature selected by the pitching motion feature selection means, it is determined whether or not the motion feature extracted by the motion feature extraction means at the media time T is a pitching motion. A determination means for outputting an event of the pitching operation;
A pitching event detection device comprising: The pitch event candidate acquisition means includes:
When acquiring the pitching event candidate, sudden sound is used,
The pitching event detection device according to claim 10. The operation feature extraction means includes
As the operation feature, Motion History Image is used.
The pitching event detection device according to claim 10 or 11. The in-section pitching motion feature acquisition means includes:
The pitching event detection device according to claim 12, wherein the motion history image of the motion feature of the motion feature extraction unit is subjected to reduction / enlargement / labeling processing to create a motion history image for each object. The pitching operation feature selection means includes:
The pitching event detection device according to any one of claims 10 to 13, which uses not only the current media time but also a pitching operation characteristic at a media time in the vicinity thereof. The limited section setting means includes
When setting the limited section, the number of pitching event candidates is used.
The pitching event detection device according to claim 10 to 14. The limited section setting means includes
When setting the limited section, a fixed media time interval is used.
The pitching event detection device according to claim 10 to 14. The limited section setting means includes
When moving the limited section, shift by a certain number of pitching event candidates,
The pitching event detection device according to claim 10 to 14. The limited section setting means includes
When moving the limited section, shift by a certain media time,
The pitching event detection device according to claim 16. On the computer,
19. A pitching event detection program that causes each means of the pitching event detection device according to claim 10 to be executed. On the computer,
A computer-readable recording medium storing a pitching event detection program for executing each means of the pitching event detection device according to claim 10.