JP6441570B2 - Golf swing analysis system, program and method - Google Patents

Description

  The present invention relates to a golf swing analysis system, program, and method for extracting addresses and / or impact frames from a moving image of a golf swing.

  Conventionally, a device for analyzing a golf swing based on a moving image of a golf swing taken by a golfer is known (Patent Documents 1, 2, etc.). The results of the analysis are used in a variety of applications such as improving golfer foam and developing golf equipment.

  By the way, the golf swing generally proceeds in the order of address, arm level, top, turn-back, impact, and finish. The address is an initial state in which the golf club head is placed near the ball, and the golf club is taken back from the address. During takeback, the state where the left arm is horizontal is horizontal, and the state where the head is swung up is the top. After the top, a downswing is started. The turn-back is the starting point of this downswing, and the golf club is swung down from the turn-back. Then, the state at the moment when the head collides with the ball is the impact, and the finish is the state in which the golf club is shaken forward after the impact.

  Positions such as address, arm level, top, turn-back, impact, and finish as described above serve as check points for diagnosing a golf swing. Therefore, it is important to automatically extract a frame at such a specific position from a moving image, and various algorithms have been proposed so far.

  Specifically, in Patent Document 1, first, an edge is detected from each frame included in a moving image, each edge image is binarized, and each binarized image is subjected to a Hough transform to correspond to a golf club shaft. Extract a straight line. An address frame is extracted by paying attention to the movement of the tip of the shaft.

  Thereafter, a frame after a predetermined number of frames from this address frame is extracted as a reference frame. The reference frame at this time is a frame in which the golf club is not captured in a partial region where the periphery of the golf club head is captured at the time of addressing. Then, an inter-frame difference value is calculated in the partial area between the reference frame and each frame after the reference frame, and a frame having the maximum inter-frame difference value is determined as an impact frame. Is done.

  Also, in Patent Document 2, a difference value between adjacent frames is sequentially calculated, and a frame at a timing at which the inter-frame difference value is minimized is determined as an address frame.

JP 2011-177341 A Japanese Patent Laying-Open No. 2005-210666

  As described above, the method of Patent Document 1 has a problem that the calculation load increases because processing such as edge image generation and Hough transform is performed on a large number of frames. However, after extracting the address frame, the process of extracting the impact frame is simply to calculate the inter-frame difference value sequentially and detect the maximum value from these, so the calculation load is relatively small. However, in the method of finding one point where the inter-frame difference value is maximum and determining that this corresponds to the impact, erroneous determination is likely to occur due to the influence of noise or the like. The same can be said for Patent Document 2 in which one point at which the inter-frame difference value is minimized corresponds to the address.

  It is an object of the present invention to provide a golf swing analysis system, program, and method capable of accurately extracting an address and / or an impact frame while suppressing a calculation load.

  A golf swing analysis system according to a first aspect of the present invention is a system that extracts an address and / or impact frame from a moving image of shooting a golf swing, and sequentially calculates a difference value between frames from the moving image. A derivation unit, a feature detection unit that detects a section in which the inter-frame difference value appears characteristically along a time axis, and a specification unit that identifies a frame of an impact and / or an address based on the detection result Prepare.

  A golf swing analysis system according to a second aspect of the present invention is the system according to the first aspect, wherein the difference derivation unit is within a partial area of the moving image in which the periphery of the head of the golf club is captured at the time of address and impact. The inter-frame difference value is sequentially derived.

  A golf swing analysis system according to a third aspect of the present invention is a system according to a second aspect, wherein the feature detection unit includes a first peak of the inter-frame difference value and a position immediately before the first peak. A section including a valley having a predetermined length or more located at is detected.

  A golf swing analysis system according to a fourth aspect of the present invention is a system according to a third aspect, wherein the feature detection unit is located immediately before the valley in addition to the first mountain and the valley. A section further including the second peak of the inter-frame difference value is detected.

  The golf swing analysis system according to a fifth aspect of the present invention is the system according to the third or fourth aspect, wherein the specific unit is selected from one or more frames corresponding to the first mountain. The frame of the impact is specified.

  The golf swing analysis system according to a sixth aspect of the present invention is the system according to the third or fourth aspect, wherein the specific unit is one or a plurality of frames corresponding to the first mountain or the valley. The frame of the address is specified from among the frames.

  A golf swing analysis system according to a seventh aspect of the present invention is the system according to any one of the first to sixth aspects, wherein the moving image is calculated before the difference value between the frames is derived by the difference deriving unit. And an image blur correction unit that corrects the image blur.

  A golf swing analysis program according to an eighth aspect of the present invention is a program for extracting an address and / or impact frame from a moving picture of a golf swing, and sequentially derives inter-frame difference values from the moving picture. And a step of detecting a section in which the inter-frame difference value characteristically appears along a time axis, and a step of identifying an impact and / or address frame based on the detection result. Let

  A golf swing analysis method according to a ninth aspect of the present invention is a method for extracting an address and / or impact frame from a moving picture of a golf swing, and sequentially derives inter-frame difference values from the moving picture. A step of detecting a section in which the inter-frame difference value characteristically appears along a time axis, and a step of identifying a frame of an impact and / or an address based on the detection result.

  In the present invention, in order to specify an address and / or an impact frame included in a moving image, inter-frame difference values are sequentially derived based on the moving image, and these values appear in a characteristic appearance pattern along the time axis. A section is detected. That is, instead of simply finding a characteristic point (for example, a point where the difference value between frames is maximum or minimum), a characteristic interval is found, and based on this, an address and / or impact frame is specified. The As a result, the possibility of erroneous determination due to the influence of noise or the like is reduced. Therefore, according to the present invention, it is possible to accurately extract an address and / or impact frame while suppressing a calculation load.

1 is a block diagram showing a configuration of a golf swing analysis system according to an embodiment of the present invention. The sequence diagram which shows the flow of a golf swing analysis process. The figure which shows an imaging | photography screen. The figure which shows the example of the flame | frame which caught the address, the arm level, the top, the cutback, the impact, and the finish. 10 is a flowchart showing a flow of checkpoint frame extraction processing. The flowchart which shows the flow of the extraction process of an impact frame. The figure which shows the graph of the difference value between frames along a time-axis. The figure which shows another graph of the difference value between frames along a time-axis. 6 is a flowchart showing a flow of address frame extraction processing. The enlarged view of the graph of the difference value between frames near an address.

  Hereinafter, a golf swing analysis system, program, and method according to an embodiment of the present invention will be described with reference to the drawings.

<1. Outline of golf swing analysis system>
FIG. 1 shows a golf swing analysis system 100 according to this embodiment. The analysis system 100 is a system for analyzing a golf swing by a golfer by photographing a golf swing as a moving image and processing the image. The analysis system 100 mainly includes a mobile phone 1 and a server 2 that can be connected from the mobile phone 1 via the Internet 6. The mobile phone 1 is equipped with a camera 10. The user takes a golf swing as a moving image using the camera 10, and the server 2 receives this from the mobile phone 1 and analyzes it. In addition, a user here is a general term for a golfer, an instructor, etc. which require the analysis result of a golf swing. Hereinafter, after describing the hardware configurations of the mobile phone 1 and the server 2, their operations will be described.

<2. Mobile phone hardware configuration>
The mobile phone 1 in the present embodiment is a smartphone with a camera function, and a program 1A for golf swing analysis is installed. This program 1 </ b> A is application software for causing the mobile phone 1 to perform an operation described later while communicating with the server 2, and can be downloaded via the Internet 6.

  The mobile phone 1 includes a display unit 11, an input unit 12, a storage unit 13, a control unit 14, and a communication unit 15 in addition to the camera 10, and these units 10 to 15 are connected via a bus line 16. They can communicate with each other. The display unit 11 is configured by a liquid crystal display or the like, and displays a screen or the like described later to the user. The input unit 12 includes a touch panel, operation buttons, and the like, and receives an operation from the user. The storage unit 13 includes a built-in flash memory, a removable SD card, and the like, and the above-described program 1A is stored here. The communication unit 15 enables voice communication with other telephones via a mobile phone communication network, data communication with various servers (including the server 2) via the Internet 6, and a dedicated cable. Communication with other information terminals is also possible. The control unit 14 includes a CPU, a ROM, a RAM, and the like, and operates as a photographing guide unit 14A and a size adjustment unit 14B virtually by reading and executing the program 1A stored in the storage unit 13. To do. The operation of each part 14A, 14B will be described later.

<3. Server hardware configuration>
The server 2 is an application server in which a program 2A for golf swing analysis is installed. This program 2A is application software for causing the server 2 to execute an operation described later while communicating with the mobile phone 1, and can be installed from a computer-readable recording medium 20 such as a CD-ROM.

  The server 2 includes a display unit 21, an input unit 22, a storage unit 23, a control unit 24, and a communication unit 25, and these units 21 to 25 can communicate with each other via a bus line 26. . The display unit 21 is configured with a liquid crystal display or the like, and displays a necessary screen or the like to the user. The input unit 22 includes a mouse, a keyboard, a touch panel, and the like, and receives an operation from the user. The storage unit 23 is configured by a hard disk or the like, and the above-described program 2A is stored here. The communication unit 25 enables data communication with other information terminals (including the mobile phone 1) via the Internet 6. The control unit 24 includes a CPU, a ROM, a RAM, and the like. The control unit 24 reads and executes the program 2A stored in the storage unit 23 to virtually execute the image blur correction unit 24A, the difference derivation unit 24B, the feature detection unit 24C, the specification unit 24D, and the diagnosis unit 24E. Works as. The operation of each unit 24A to 24E will be described later.

<4. Golf swing analysis processing>
Hereinafter, the golf swing analysis process by the mobile phone 1 and the server 2 will be described with reference to FIG.
First, when a user wants to improve a golf swing at a driving range or the like, the user activates a program 1A for supporting this on the mobile phone 1 (step S1). In response to this, the shooting guide unit 14A of the mobile phone 1 displays a shooting screen W1 as shown in FIG. 3 on the display unit 11 (step S2). A through image captured by the camera 10 is displayed in the photographing screen W1, and two guide frames W2 and W3 are displayed on the through image. The center guide frame W2 is a frame indicating a region where the grip 71 of the golf club 7 gripped by the golfer 70 is to be captured at the time of addressing, and the lower right guide frame W3 is the head 72 and the ball of the golf club 7 at the time of addressing. This is a frame indicating the area to be captured. Therefore, the user can easily understand the approximate position and angle at which the golf player 70 should be captured within the imaging range of the camera 10 by referring to the guide frames W2 and W3. The guide frames W2 and W3 guide the golfer 70 to be captured at a predetermined position (in the present embodiment, near the center) within the imaging range of the camera 10 while holding the camera 10 vertically long. On the shooting screen W1, in addition to the guide frames W2 and W3, information indicating the meaning thereof, for example, “the player's hand is copied in the center frame, and the golf club head is displayed in the lower right frame. You may see a message saying "Please copy."

  Subsequently, the user captures the golfer 70 at an appropriate position within the imaging range of the camera 10 according to the above guidance, and then shoots a moving image showing the golf swing with the camera 10 (step S3). At this time, the user captures a series of operations from the address to the finish while taking care not to move the camera 10 held in his hand as much as possible. A golf swing generally proceeds in the order of address, arm level, top, turn-back, impact, and finish. FIG. 4 shows an example of a frame that captures these six states. The address is an initial state in which the head 72 of the golf club 7 is disposed near the ball, and the golf club 7 is taken back from the address. During takeback, the state where the left arm is horizontal is the arm horizontal, and the state where the head 72 is swung up most is the top. After the top, a downswing is started. The turn-back is the starting point of this down swing, and the golf club 7 is swung down from the turn-back. Then, the state at the moment when the head 72 collides with the ball is an impact, and the finish is a state in which the golf club 7 is swung forward after the impact. The captured moving image is stored in the storage unit 13.

  When shooting is finished, the size adjustment unit 14B of the mobile phone 1 reduces the image size of the shot moving image as necessary (step S4). Specifically, the size adjustment unit 14B determines the aspect ratio of the captured moving image, and when the aspect ratio is determined to be different from a predetermined value (for example, when the aspect ratio is 16: 9), the aspect ratio is determined. An area of a predetermined percentage is cut from the edge in the vertical direction (longitudinal direction) of each frame included in the moving image so that the ratio becomes a predetermined value (in this embodiment, vertical 4: horizontal 3). This is because it is considered that the golfer 70 and the golf club 7 are not captured in a predetermined% area above and below each frame when the image size is vertically long at a ratio of a certain level or more. Further, the size adjustment unit 14B determines the number of pixels of the above moving image, and when it is determined that this is larger than a predetermined value (for example, vertical 1440 × horizontal 1080), In the embodiment, each frame included in the moving image is compressed so that the length is 640 × width 480). At this time, it is preferable to convert the pixel values into integers for each color component (R, G, B) by interpolation processing so that no decimal point is generated in the compressed pixel values.

  Subsequently, the communication unit 15 of the mobile phone 1 transmits the above moving image to the server 2 via the Internet 6 (step S5). On the other hand, on the server 2 side, the communication unit 25 receives this, and then the image blur correction unit 24A performs image blur correction on the moving image sent from the mobile phone 1 (step S6). Since various image blur correction algorithms are known, detailed description thereof is omitted here. By this image blur correction, image blur between frames included in the moving image is eliminated.

  When the image blur correction of the moving image is completed, the difference deriving unit 24B, the feature detecting unit 24C, and the specifying unit 24D of the server 2 cooperate to automatically select a frame at a specific timing from the moving image after the image blur correction. Extract (step S7). Specifically, six frames of address, arm horizontal, top, cut-back, impact and finish are automatically extracted. Such timing becomes a check point for diagnosing a golf swing. Details of the check point frame extraction processing will be described later.

  Thereafter, the diagnosis unit 24E diagnoses a golf swing using the checkpoint frame (step S8). For example, the silhouette of the golfer 70 or golf club 7 reflected in the frame of the checkpoint is automatically extracted, and the arm is not bent where it should be stretched, the face is facing the correct direction, the golf club is oriented correctly, etc. A check item is diagnosed. Such check items can be created based on the knowledge of the instructor. It is also possible to prepare a frame at a checkpoint when a professional performs a golf swing and compare it with a frame at a checkpoint by the user. Note that various algorithms for diagnosing golf swings are known, and thus detailed description thereof is omitted here.

  The result of the above diagnosis is transmitted from the communication unit 25 of the server 2 to the communication unit 15 of the mobile phone 1 (step S9) and displayed on the display unit 11 of the mobile phone 1 (step S10). Therefore, the user can confirm the result of the diagnosis and use it for learning the correct swing form. The result of the diagnosis can be expressed as a swing score, advantages / disadvantages, a practice menu corresponding to the golfer 70, and the like. Also, the checkpoint frame itself may be the result of diagnosis.

<5. Checkpoint frame extraction processing>
Next, a checkpoint frame extraction process will be described with reference to FIG. In the following, a plurality of frames included in the moving image sent from the mobile phone 1 and subjected to the above-described image blur correction are sequentially F ₁ , F ₂ ,..., F _n along the time axis. (N is the number of frames). In the present process, first, the impact frame F _I from the video is extracted (step S21), and subsequently, address frame F _A is extracted (step S22). The extraction process of the impact frame F _I and the address frame F _A will be described later.

Next, the top frame _FT is extracted from the moving image (step S23). Here, the top frame _FT is searched based on the timing of impact. Specifically, top, to produce before the impact, the impact frame F N2 immediately previous frame from N1 immediately previous _I F _I-N1, ···, between F _I-N2, the top frame F _T is Explored. If the FPS (Frame per second) of the moving image is 30 fps, for example, N1 = 15 and N2 = 4.

As a search algorithm, the difference deriving unit 24B sequentially executes difference processing between the adjacent frames F _i-1 and F _{i in the} search period i = (I−N1) to (I−N2). The difference process is a process for calculating an inter-frame difference value. In the present embodiment, an absolute difference sum (SAD) is calculated as the inter-frame difference value. That is, for each pixel, the difference deriving unit 24B creates a difference image using a value obtained by subtracting the pixel value in the frame F _i-1 from the pixel value in the frame F _{i as} the pixel value of the pixel. Here, the calculation of the pixel value is performed for each color component (R, G, B). Then, the difference deriving unit 24B calculates a total value (SAD) of absolute values of pixel values regarding all color components of all pixels included in the difference image. Specifying unit 24D, after the difference processing in the search period i = (I-N1) ~ (I-N2), to identify i = T the difference value between frames is minimized, the top frame F _T at this time Frame. Note that the difference processing in step S23 is executed for the entire moving image screen.

Next, the arm horizontal frame F _H is extracted from the moving image (step S24). Here, the arm horizontal frame F _H is searched with reference to the timing of the address. More specifically, since the arm level occurs after the address, the arm level frame F _H between the frames F _{A + N3} ,..., F _{A + N4 N3} frames to N4 frames after the address frame F _A. Is searched. If the FPS of the moving image is 30 fps, for example, N3 = 10 and N4 = 45 can be set.

As a search algorithm, the difference deriving unit 24B sequentially creates a difference image between the frame F _i and the address frame F _{A in the} search period i = (A + N3) to (A + N4). This difference image can be created by various methods. For example, for each pixel, the distances between F _i and F _A relating to hue, saturation and brightness in the HSV color space are appropriately weighted and added. An image having the obtained value (hereinafter referred to as a color distance) as the pixel value of the pixel is generated, and this can be further binarized. In this binarization, the pixel value of a pixel having a color distance larger than a predetermined threshold is set to 1 (black), and the pixel values of other pixels are set to 0 (white). At this time, in order to reduce the influence of noise, when the area where black continues is less than a predetermined size, the pixel value of the pixel in the area can be converted to 0 (white).

Subsequently, the feature detection unit 24C sequentially detects a straight line corresponding to the shaft 73 of the golf club 7 on the above difference image by the Hough transform in the search period i = (A + N3) to (A + N4). In the difference image between the frame F _{i at the} search time and the address frame F _A , the shaft 73 at the address and the shaft 73 at the search time are shown. As can be seen from FIG. 4, when a golf swing is photographed from the rear with respect to the flight direction of the ball, the straight line of the shaft 73 when the arm is horizontal is longer than the shaft 73 at the time of addressing, and at least The search period here has the maximum length. Accordingly, the specifying unit 24D specifies i = H where the straight line that has obtained the maximum number of votes in the search period i = (A + N3) to (A + N4), that is, the straight line corresponding to the shaft 73 when the arm is horizontal exists. The frame F _H at this time is referred to as an arm horizontal frame F _H. Since the angle of the shaft 73 when the arm is horizontal is determined to some extent, the parameter θ representing the straight line is set within a predetermined range (for example, −5 ° to −85 °) when searching for the straight line. It is preferable.

Next, the specification unit 24D, based on the timing of the impact, to extract the forward turning frame F _S from the video (step S25), and further extracts the finish frame F _F (step S26). Specifically, the specifying unit 24D includes a frame F _S crosscut the N5 immediately previous frame of the impact frame F _I, the N6 sheets after the frame of the impact frame F _I and finish the frame F _F. If the FPS of the moving image is 30 fps, for example, N5 = 3 and N6 = 15 can be set.

<5-1. Impact Frame Extraction Processing>
Next, referring to FIG. 6, it will be described extraction process of the impact frame F _I. Hereinafter, the partial image in the guide frame W3 in the frame F _i, expressed as G _i. The partial image G _i, as described above, an image of a region Utsuru the periphery of the head 72 and the ball of the golf club 7 at address and impact.

First, the difference deriving unit 24B sequentially selects adjacent frames in the area within the guide frame W3 with respect to i = 2, 3,..., N−N7 (for example, N7 = 10 if the FPS of the moving image is 30 fps). Difference processing is executed between F _i-1 and F _i (step S31). That is, the difference deriving unit 24B sequentially performs the difference process between the partial images G _i−1 and G _{i for} i = 2, 3,..., N−N7. The reason why the difference processing for the section from i = (n−N7 + 1) to n is omitted is that there is no impact in such a period. The difference process in the present embodiment is a process for calculating a sum of absolute differences (SAD) as an inter-frame difference value as described above. Accordingly, here, the difference deriving unit 24B, for each pixel, to create a difference image to the value obtained by subtracting the pixel values in the partial image G _i-1 from the pixel values in the partial image G _i and the pixel value of the pixel. Here, the calculation of the pixel value is performed for each color component (R, G, B). Then, the difference deriving unit 24B calculates a total value (SAD) of absolute values of pixel values regarding all color components of all pixels included in the difference image. Therefore, in step S31, SAD (2), SAD (3),... SAD (n−N7) are obtained. SAD (i) is SAD between the partial images G _i-1 and G _i .

By the way, the SAD calculated in step S31 is an index representing a change occurring in the guide frame W3 during a very short time. Further, as described above, the guide frame W3 is a frame that surrounds an area that captures the periphery of the head 72 of the golf club 7 and the ball at the time of addressing and impact. Therefore, the partial image G _i in the guide frame W3 is mainly in the non-short period before and after the address and impact, there is not much change. Therefore, the SAD value calculated in step S31 is expected to draw a graph as shown in FIG. 7 in which an impact peak 51 and an address peak 53 appear. However, often, due to the movement of a ball, a tee or the like, as shown in FIG. However, in such a case, the width of the valley 55 between the impact peak 51 and the noise peak 54 is shorter than the width of the valley 52 between the impact peak 51 and the address peak 53. .

Therefore, in subsequent steps S32 to S37, the feature detection unit 24C detects a section in which the value of the SAD calculated in step S31 characteristically appears in the appearance pattern as described above along the time axis. Specifically, the feature detection unit 24C includes a first peak of SAD equal to or greater than a predetermined threshold D ₁ (D ₁ = 5000 in the present embodiment), and a predetermined threshold D positioned immediately before the first peak. SAD valleys less than ₁ are detected. In this embodiment, the width of the mountain is not particularly limited, but the valley has a width equal to or greater than a predetermined length D ₂ (for example, if FPS of a moving image is 30 fps, D ₂ = 10 frames). What it has becomes a detection target. Here, the appearance pattern of the SAD having such characteristics is detected, and the first peak is determined as the impact peak 51, so that the noise peak 54 after the impact is erroneously detected as the impact peak 51. Can be prevented.

  The above appearance patterns are detected by scanning the values of SAD (2), SAD (3),... SAD (n−N7) in reverse feed along the time axis. The feature detection unit 24C substitutes n−N7 for the parameter i in order to set the scan start time (step S32).

Then, feature detecting section 24C until the threshold D ₁ or more SAD are detected, Yuku sequentially checks the value of the previous SAD along the time axis (step S33). When the threshold value D ₁ or more SAD is detected, the feature detection unit 24C is in turn until small SAD is detected than the threshold value D _1, sequentially checks the value of the previous SAD along the time axis Go (step S34). Therefore, in step S33, S34, mountain threshold D ₁ or more SAD are continuous are detected. This mountain is a candidate for the impact mountain 51.

When smaller SAD than the threshold value D ₁ in step S34 is detected, i.e., when the beginning of the end of the peaks and valleys are detected, wherein the detection unit 24C is the initial value to the counter cnt for measuring the width of the trough 1 is set (step S35). The feature detection unit 24C uses the counter cnt, determines whether the valley has a length D ₂ or the width (step S36, S37). Specifically, in step S36, as long as the smaller SAD than the threshold value D ₁ is detected, Yuku sequentially checks the value of the previous SAD along the time axis. At this time, the feature detection unit 24C while incrementing the value of cnt, counts the width of the valley smaller SAD than the threshold D ₁ are continuous.

The feature detecting section 24C is smaller SAD than the threshold value D ₁ is detected in step S36, each time the value of cnt is incremented, the value of the current cnt, i.e., checks the width of the root (step S37) . Then, feature detecting section 24C, when the width of the valley is determined to be a predetermined length D ₂ or more, the detected as a candidate of a mountain 51 impact at step S34 immediately before the mountain, finally Is determined as the impact mountain 51. Then, feature detecting section 24C, from among the one or more frames corresponding to the pile 51 of the impact, to identify the impact frame F _I (step S38). Specifically, in the mountains that are detected at step S34 immediately before to identify i = I where the value of SAD (i) becomes the maximum, the frame F _I at this time is an impact frame.

On the other hand, the value of the case (counter cnt to the width of the root in step S36, S37 is determined to less than a predetermined length D ₂ is before reaching the D _2, it is detected threshold D ₁ or more SAD is, the valleys When the end is detected, the feature detection unit 24C returns the process to step S34. That is, the mountain detected as the candidate for the impact mountain 51 in the immediately preceding step S34 is finally determined as the noise mountain 54, and the same scanning is continued. As can be seen from FIG. 6, in this process, any number of noise peaks 54 after the impact peak 51 can be detected as noise peaks.

<5-2. Address frame extraction processing>
Next, referring to FIG. 9 will be described extraction process of the address frame F _A. In this process, the feature detection unit 24C uses the SAD values calculated in step S31 to detect a section in which these SAD values characteristically appear in a predetermined appearance pattern along the time axis.

  More specifically, although not shown in FIG. 7, the address peak 53 may be composed of a plurality of peaks as shown in FIG. This is because the golfer swings the golf club 7 left and right for positioning before the address, and also performs a waggle operation that swings the golf club 7 left and right after the address. However, at the true address, the golfer rests for a short time. Accordingly, the portion corresponding to the address on the SAD graph is expected to have a shape in which peaks 532 to 534 appear at least before and after the valley 531 having a predetermined length or more (at least the mountain 533 after the valley 531). The

Therefore, the feature detection unit 24C has a first peak of SAD that is equal to or greater than a predetermined threshold value D ₃ (in this embodiment, D ₃ = 5000 as in the case of the threshold value D ₁ ), and a predetermined value positioned immediately before the first peak. And a SAD valley that is smaller than the threshold D ₃ and a second SAD peak that is not less than a predetermined threshold D ₃ and is located immediately before the valley. In this embodiment, the width of the mountain is not particularly limited, but the valley has a width equal to or greater than a predetermined length D ₄ (for example, if FPS of a moving image is 30 fps, D ₄ = 5 frames). What it has becomes a detection target. Here, the appearance pattern of the SAD having such characteristics is detected, and the first peak is determined as the true address peak 533, so that the noise peak 534 is erroneously detected as the true address peak 533. Can be prevented.

  The above appearance patterns are obtained by reversing the values of SAD (2), SAD (3),... SAD (I-N8) (for example, N8 = 5 if FPS = 30 fps of the moving image) along the time axis. It is detected by scanning with feed. The reason why the difference processing for the period from i = (I−N8 + 1) to n is omitted is that there is no address in such a period. The feature detection unit 24C substitutes I-N8 for the parameter i in order to set the scan start time (step S41).

Then, feature detecting section 24C until the threshold D ₃ or more SAD are detected, Yuku sequentially checks the value of the previous SAD along the time axis (step S42). When the threshold value D ₃ or more SAD is detected, the feature detection unit 24C is in turn until small SAD is detected than the threshold value D _3, sequentially checks the value of the previous SAD along the time axis Go (step S43). Therefore, in step S42, S43, mountain threshold D ₃ or more SAD are continuous are detected. This mountain is a candidate for the true address mountain 533. Note that the feature detection unit 24C specifies the first frame F ₁ as the address frame F _A when the target SAD is not detected even after examining SAD (2) in steps S42 and S43 (step S48). ).

When smaller SAD than the threshold value D ₃ in step S43 is detected, i.e., when the beginning of the end of the peaks and valleys are detected, wherein the detection unit 24C is the initial value to the counter cnt for measuring the width of the trough 1 is set (step S44). Then, feature detecting section 24C until the next time threshold D ₃ or more SAD are detected, Yuku sequentially checks the value of the previous SAD along the time axis (step S45). At this time, the feature detection unit 24C while incrementing the value of cnt, counts the width of the valley smaller SAD than the threshold value D ₃ are continuous.

When an SAD greater than or equal to the threshold value D ₃ is detected in step S45, that is, when the end of the valley and the start of a new mountain are detected, the feature detection unit 24C determines the current cnt value, The width of the valley is checked (step S46). Then, feature detecting section 24C, when the width of the valley is determined to be a predetermined length D ₄ above, were detected in the step S43 immediately before as a candidate for the true address of the mountains 533 mountains, Finally, the peak 533 of the true address is determined. Subsequently, the feature detection unit 24C identifies the address frame F _A from one or more frames corresponding to the true address peak 533 (step S47). Specifically, of the value of i corresponding to the mountain detected in step S43 immediately before to identify the earliest i = A along the time axis, the frame F _A at this time it is the address frame. The feature detecting section 24C, even when the threshold value D ₃ or more SAD is not detected examine up SAD (2) in step S45, the process proceeds to step S47.

On the other hand, if the width of the immediately preceding valley is determined to less than a predetermined length D ₄ in step S46, the feature detection unit 24C, the process returns to step S43. That is, the peak detected as a candidate for the true address peak 533 in the previous step S43 is finally determined as the noise peak 534, and the same scanning is continued. As can be seen from FIG. 9, in this process, any number of noise peaks 534 after the true address peak 533 can be detected as noise peaks.

<6. Features>
In the above embodiment, in order to identify the address frame F _A and the impact frame F _I included in the moving image, inter-frame difference values (SAD) are sequentially derived based on the moving image, and these values are characteristic along the time axis. Sections that appear in various appearance patterns are detected. That is, instead of simply finding a characteristic point (for example, a point where the inter-frame difference value is maximum or minimum), a characteristic interval is found, and based on this, the address frame F _A and the impact frame F _I are determined. Identified. As a result, the possibility of erroneous determination due to the influence of noise or the like is reduced. Therefore, according to the above embodiment, the address frame F _A and the impact frame F _I can be extracted with high accuracy while suppressing the calculation load.

<7. Modification>
As mentioned above, although one Embodiment of this invention was described, this invention is not limited to the said embodiment, A various change is possible unless it deviates from the meaning. For example, the following changes can be made. Moreover, the gist of the following modifications can be combined as appropriate.

<7-1>
In the above embodiment, shooting of the golf swing and reduction of the image size are executed on the mobile phone 1 side, and image blur correction, checkpoint frame extraction, and golf swing diagnosis are executed on the server 2 side. It was. However, the main body that executes these processes can be appropriately assigned to the mobile phone 1 and the server 2. For example, the mobile phone 1 may be configured to execute from shooting to image blur correction, or after only shooting with the mobile phone 1, the video before reduction is transmitted from the mobile phone 1 to the server 2, The image size may be reduced on the server 2 side. Alternatively, the server 2 may be omitted and the cellular phone 1 may execute all the processes from photographing to diagnostic processing.

<7-2>
The hardware configuration of the analysis system 100 is not limited to the above. For example, the image may be taken by a camera other than a camera having an Internet communication function such as a mobile phone, and this may be taken into a personal computer (hereinafter referred to as a PC) and transmitted from the PC to the server 2. Alternatively, a program similar to that of the server 2 may be installed on the PC and analyzed by the PC.

<7-3>
When the processing capability of the cellular phone 1 or the server 2 is sufficiently high, the upper and lower area cutout processing and / or compression processing in step S4 can be omitted as appropriate.

<7-4>
In the above embodiment, the SAD is calculated as the inter-frame difference value, but another index can be used as long as the inter-frame difference can be appropriately evaluated. For example, a sum of squared differences (SSD) may be calculated as the inter-frame difference value. Or you may calculate the pixel number from which the absolute value of a pixel value becomes more than predetermined value in a difference image.

<7-5>
In the above embodiment, at the time of impact detection, first peak, all the threshold for detecting a valley and second peak was the same value D _1, may be these thresholds as appropriate different values. The same applies to the threshold D ₃ at the time of address detection. Further, D ₁ may not be D ₃ .

<7-6>
In the above-described embodiment, scanning is performed in reverse to detect the appearance pattern of SAD at the time of impact and address detection. However, the scanning order is not limited to this, and for example, scanning may be performed from the front.

<7-7>
In the above embodiment, when the address is detected, the first peak is determined to be a true address peak because the second peak is detected after the valley having a predetermined length or more. However, the detection of the second peak is omitted, and after the first peak, a valley having a predetermined length or more is detected, so that the first peak is determined to be a true address peak. Also good. Further, when the impact is detected, the first mountain may be determined as the impact mountain because the second mountain is detected after the valley having a predetermined length or more.

<7-8>
In the above embodiment, the first frame in the frame group corresponding to the peak 533 of the true address is the address frame F _A. However, the address frame F _A is detected from another part of the peak 533 of the true address. Alternatively, the frame corresponding to the valley 531 may be the address frame F _A. Further, at a point where SAD (i) is not a maximum of impact mountain 51 may detect the impact frame F _I.

<7-9>
In step S31 of the above embodiment, the difference process may be executed between the frame at the search time and the predetermined reference frame instead of between adjacent frames. The reference frame at this time can be, for example, an image at a timing when the golf club 7 is not shown in the guide frame W3.

1 Golf Swing Analysis System 1A, 2A Program (Golf Swing Analysis Program)
24A Image blur correction unit 24B Difference derivation unit 24C Feature detection unit 24D Identification unit 51 Mountain of impact (first mountain)
52 Valley 53 Mountain of Address (Second Mountain)
531 Valley 532 Mountain (second mountain)
533 True address mountain (first mountain)
7 golf club 72 head

Claims (5)

A golf swing analysis system that extracts an address frame from a video of shooting a golf swing,
In the partial area of the moving image in which the periphery of the head of the golf club is reflected from the moving image at the time of impact and impact, a difference deriving unit that sequentially derives a difference value between frames
A section in which the inter-frame difference value appears characteristically along the time axis, and a first peak of the inter-frame difference value and a valley of a predetermined length or more located immediately before the first peak A feature detection unit for detecting a section including
A golf swing analysis system comprising: a specifying unit that specifies a frame of an address from one or a plurality of frames corresponding to the first mountain or the valley based on the detection result. The feature detection unit detects a section further including a second peak of the inter-frame difference value located immediately before the valley in addition to the first peak and the valley.
The golf swing analysis system according to claim 1 . An image blur correction unit that corrects image blur of the moving image before the inter-frame difference value is derived by the difference deriving unit;
The golf swing analysis system according to claim 1 or 2 . A golf swing analysis program for extracting an address frame from a video of shooting a golf swing,
From the moving image, in the partial area of the moving image in which the periphery of the head of the golf club is captured at the time of address and impact, sequentially deriving inter-frame difference values;
A section in which the inter-frame difference value appears characteristically along the time axis, and a first peak of the inter-frame difference value and a valley of a predetermined length or more located immediately before the first peak Detecting a section including
A golf swing analysis program that causes a computer to execute a step of identifying a frame of an address from one or a plurality of frames corresponding to the first mountain or the valley based on the detection result. A golf swing analysis method for extracting a frame of an address from a video of shooting a golf swing,
From the moving image, in the partial area of the moving image in which the periphery of the head of the golf club is captured at the time of address and impact, sequentially deriving inter-frame difference values;
A section in which the inter-frame difference value appears characteristically along the time axis, and a first peak of the inter-frame difference value and a valley of a predetermined length or more located immediately before the first peak Detecting a section including
A method of analyzing a golf swing, comprising: specifying a frame of an address from one or a plurality of frames corresponding to the first mountain or the valley based on the detection result.