AU2015246642B2

AU2015246642B2 - Sports throwing measurement

Info

Publication number: AU2015246642B2
Application number: AU2015246642A
Authority: AU
Inventors: Glenn MCINTOSH
Original assignee: Catapult Group International Pty Ltd
Current assignee: Catapult Group International Pty Ltd
Priority date: 2014-04-18
Filing date: 2015-04-17
Publication date: 2017-06-29
Anticipated expiration: 2035-04-17
Also published as: GB2544150B; WO2015157808A1; GB2544150A; AU2015246642A1

Abstract

A data collection system includes a data collection device mounted on the upper torso of the player. The device includes a GPS unit, three dimensional accelerometers, three dimensional gyroscopes, and a microprocessor and data store for processing the signals from the GPS unit, three dimensional accelerometers and three dimensional gyroscopes. The microprocessor is programmed to identify and collect signal data to derive delivery rotation speed, point of release and to derive delivery speed and maintain a count of deliveries and optionally the distance and speed of the player during the act off delivery. When the software is adapted for cricket it derives bowling count, delivery speed, run up velocity and run up distance. It may also be adapted for baseball pitching and tennis serving where the device is worn on the forearm or wrist.

Description

PCT/AU2015/000231 wo 2015/157808

SPORTS THROWING MEASUREMENT

This invention relates to a system and software for collecting and analysing data relating to ball throwing for such sports as cricket, baseball and tennis serves. 5 Background to the invention

In sports such as cricket, baseball and tennis a lot of training time is spent practising bowling, throwing, pitching and serving. The analysis of the players action in bowling, throwing, pitching and serving is an essential part of developing skill and avoiding stress injuries at the elite level. Maintaining a record of training drills is also 10 essential to good player management.

An article in the journal of sports science August 2000 by Paul Glazier discloses a system using two cameras and a radar gun to measure delivery speed of cricket fast bowlers and analyse back foot impact, front foot impact, ball release and follow through. 15 USA patent 5419549 provides a system for coaching baseball pitchers which uses a target with impact sensors to locate impact points and spin of the ball and a radar gun is used to measure speed. USA patent 5566934 discloses a baseball pitching practice system with a target system and a pitching mat to sense ball release when the pitchers back foot is 20 raised. The time of the two events enables ball speed to be measured. USA patent 6079269 measures ball speed by incorporating a miniature radar sensor in a catching glove. USA patent 8142267 discloses a system for training baseball pitchers which uses as data capture devices one or more video cameras and motion markers located at 25 movement points on the pitchers body. The software uses the data to provide an output that may be compared to a pitch plan. USA patents 7715982 and 8036826 disclose a sports data logger for sports which includes accelerometers, gyro scopes and GPS units for collecting sports related statistics. 30 USA patent application 20130150121 discloses a software program for a mobile phone that incorporates accelerometers and gyrometers. The user moves the phone to simulate a sports motion such as a golf club or baseball pitch and the software. PCT/AU2015/000231 wo 2015/157808 analyses the movement and provides appropriate content from a database to assist in improving the movement.

It is an object of this invention to provide a system and software for collecting data relevant to bowling and throwing in cricket, throwing and pitching in baseball and serving in tennis.

Brief description of the invention

To this end the present invention provides a data collection system which includes a data collection device mounted on the upper torso of the player 10 said device including a GPS unit, three dimensional accelerometers, three dimensional gyroscopes, and a microprocessor and data store for processing the signals from the GPS unit, three dimensional accelerometers and three dimensional gyroscopes said microprocessor being programmed to identify and collect signal data to derive 15 one or more of delivery rotation speed, point of release and to derive delivery speed and maintain a count of deliveries and optionally the distance and speed of the player during the act off delivery.

For cricket the preferred parameters of interest are the detection/categorizing of delivery actions, measuring the acceleration velocity and length of the delivery stride 20 (using the inertial sensors in combination), measuring the delivery speed (again calculated by combining rotation and velocity inputs), and the orientation of the limb or torso at the extremes of travel. Orientation may be calculated by combining the accelerometers and gyro using a Kalman filter.

The device preferably consists of a back mounted unit with inertial sensors, 25 accelerometers measuring 3D acceleration and gyroscopes measuring 3D rotational velocity of the upper torso. There is also a GPS providing velocity from doppler measurements. A unit of this kind is described in the applicants USA patent 8036826 the content of which is incorporated herein by reference.

Several parameters are derived from the signal out puts of the various sensors. 30 The delivery rotation speed is derived using the sum of squares routine from the roll and yaw rotation of the upper torso.

The delivery is detected by looking for a significant peak in the rotation speed (preferably greater than 5007s). Preferably, the centre of this peak is assumed to be PCT/AU2015/000231 wo 2015/157808 the point of release. Several other data characteristics are preferably used to make this specific to the cricket bowling action; the algorithm uses the back-foot/coil contact, looking for an upwards acceleration prior to the rotation peak (greater than 4g within 0.1s prior); 5 the algorithm checks that the rotation has a strong roll component (roll greater than 2/3 of the yaw): and the algorithm checks that the run-up distance is sufficient (greater than 2.5m). The run up distance is calculated by integrating the velocity from the point at which the speed exceeds 1.5 m/s up until the rotation peak. 10 The delivery speed is estimated from the forward velocity (GPS) at delivery together with the product of the arm span and the rotational velocity peak (gyroscopes).

The method is also applicable with some variations to other throwing and hitting actions, such as baseball deliveries or tennis serves. For those actions which involve bending of the elbow, it is preferred to measure the inertial parameters 15 (acceleration and rotational velocity) from a unit placed on the forearm or wrist.

The electronics may be mounted on the arm (preferably near the wrist) in a compression sleeve or sweatband or it may alternatively be a watch if this is allowed in the sport. The electronics assembly has a processor; a clock (real time): a battery - preferably rechargeable: accelerometers: gyroscopes: magnetometers: 20 preferably a heart rate measurement device more preferably an optical heart rate device. The wireless communications are typicaiiy ANT and /or Bluetooth. The instrumented sleeve or wristband may also include a button to enable the wearer to input some data. The electronic assembly is very small and typically weighs only 3-4 grams. In cricket and tennis the unit may be worn on the main bowling arm or 25 serving arm.

When installed on a compression sleeve the device of this arm mounted version of the invention has the additional and well documented advantages of decreasing injury, muscle soreness and increasing recovery time. 30 Detailed description of the invention A preferred embodiment of the invention will now be described with reference to the drawings in which : PCT/AU2015/000231 wo 2015/157808

Figure 1 is a schematic of the data flow from the sensors to derivation of the output parameters:

Figure 2A is a screen shot of a bowling spell during a cricket match;

Figure 2B displays a table of the data relating to figure 2A;

Figure 3A is a screen shot of a single run up and delivery;

Figure 3B displays a table of the data relating to figure 3A;

Figure 4 is a screen shot of a few steps during a bowling delivery;

Figure 5 illustrates the sensor signals during a cricket bowling delivery. 10 For the data flow diagram of figure 1 the 3 dimensional accelerometers, 3 dimensional gyrometers and the GPS unit are integrated into a device worn on the upper torso. A unit of this kind is described in the applicants USA patent 8036826. This device also includes wireless transmitters to provide a live feed of data to a coach’s computer, for display in various graphical or tabular formats. 15 As shown in figure 1 the data from the accelerometers and gyroscopes are combined using a Kalman filter to derive orientation, absolute acceleration and tilt. Using a double integration technique the height of delivery may be derived from the vertical acceleration component. Using the vertical acceleration component and a wavelet filter the foo strikes may be determined. 20 The peak detection of vertical acceleration is marked by a significant peak in the rotation speed (preferably greater than 5007s). Preferably, the centre of this peak is assumed to be the point of release. The algorithm uses the back-foot/coil contact, looking for an upwards acceleration prior to the rotation peak (greater than 4g within 0.1s prior). The algorithm checks that the rotation has a strong roll component (roll 25 greater than 2/3 of the yaw). This then is used as an indication pf delivery and is used to count each bowling delivery.

The GPS data is used to derive distance travelled. The run up distance is calculated by integrating the velocity from the point at which the speed exceeds 1.5 m/s up until the rotation peak. The algorithm checks that the run-up distance is sufficient 30 (greater than 2.5m) and this is used to conform the bowling count..

The Gyroscope data provides information for determining roll and rotation.

The acceleration velocity and length of the delivery stride is made using the inertial sensors in combination. PCT/AU2015/000231 wo 2015/157808

The delivery speed is calculated by combining rotation and velocity inputs, and the orientation of the limb or torso at the extremes of travel. The delivery speed is estimated from the fonward velocity (GPS) at delivery together with the product of the arm span and the rotational velocity peak (gyroscopes). 5 For Cricket, the parameters to be determined and displayed are; bowling count - a running count of deliveries over time; delivery speed - a running measure of the speed. This represents the delivery speed at the point where the bowling count is incremented on the rotation peak, smooth velocity . - the velocity over time, provides a measure of the bowler's run 10 up velocity run up distance - accumulates distance during the bowler's run up These parameters are collected and are able to be transmitted to the coach’s computer for analysis or for storage and later analysis. The data may be synchronised with video of the same event 15 Figure 2A is a screen shot of data from a match across a bowling spell. The automated bowling count is the top graph. The six staircase sections for each delivery in each over is shown. The delivery speed data and the smooth velocity graphs are shown in the figure 2B and synchronised with the bowling count.

Figure 3A is a screen shot of a single run up and delivery. In the bottom graph, the 20 run up velocity is continuous , and the run up distance is reset back to zero after the delivery. The peak in the middle graph is the estimated delivery speed (in this case about 106 km/h).

Figure 4 shows just the few steps around a delivery during a training session. This may be synchronized with the video that can be shown next to the graph. 25 A cricket delivery consists of several phases. The sensor signals during a bowling delivery are shown in Figure 5. The 3 dimensions of the accelerometers are shown at top and the gyrometers in the middle. The bowler’s velocity is shown at the bottom.

The run-up is characterized by increasing velocity (bottom graph), and periodic 30 vertical accelerations (top graph) of a couple of g with each stride. The vertical accelerations reach a peak at the pre-delivery stride. Immediately prior to the delivery itself, the back foot contact converts some of the forward motion into height, and this is characterized by a sharp fon^/ard deceleration (top graph centre right). PCT/AU2015/000231 wo 2015/157808

The front foot contact occurs with the body more side on, driving the rotation of the body around the back, characterized by sideways and upwards acceleration of several g. As the outstretched bowling arm comes over the top, the ball is released, characterized by a peak in the rotation (middle graph centre right) of some hundreds 5 of degrees per second. Following the release there is a recovery step or two, characterized by a side component of acceleration, before settling back into several normal steps ending the fonA/ard motion (bottom graph) characterized by the usual periodic vertical accelerations (top graph right), but with a slower cadence than the run-up. 10 In order to detect and analyse this delivery, the data logger makes use of an STM32F400 series CPU, with a GPS unit to measure triangulated position and doppler velocity at 10Hz, and MPU6000 inertial sensors to measure 3D accelerations over a range of +-15 g and rotational velocities over a range of +-1500 °/s at a rate of 10OOHz. This is placed in a vest and located on the upper back. The 15 unit also includes a real time ciock, wireless communications, and is battery powered. The inertial sensors are calibrated against reference conditions.

For bowling, the measurements are used to compute several measures: a count of the number of deliveries, the rotational magnitude of each delivery, the run-up distance of the delivery, an estimated speed of the delivery, and the run-up velocity 20 peak.

Sensor conditioning

The inertial sensors are preferably measured with 12 bit precision. This is down sampled to a rate of 100Hz after filtering using an HR filter with a gentle roll-off. Kalman Filter 25 The Kalman filter takes the 3D accelerometer and 3D gyroscope measurements at 100Hz and produces an estimate of the unit orientation. In the filter, the system state is modelled as a 3D orientation and 3D rotational velocity.

The filter is based on Kalman theory, where the state is used to estimate forward in time (predict step), and the difference between this forward estimation and the 30 inputs (the measurement residual) is used to correct the estimation (update step). The accelerometer measurements are given a low weighting as measurements of orientation, and the gyroscope measurements are given a high weighting as measurements of rotation. PCT/AU2015/000231 wo 2015/157808

The device uses an unscented variation of the Kalman filter, since orientation is not a linear field. In the unscented Kalman filter, the predict and update functions are assumed to be non-linear. A sampling technique (the unscented transform) is used to select a representative set of points, which are propagated through the predict 5 and update functions so that the mean and covariance of the estimates can be calculated statistically.

The device uses quaternions to measure rotations, and to express orientations as a rotation from the nominal zero orientation. When applying the filter, we split the rotation up into an accumulated orientation and a rotation offset from that. Only the 10 offset is part of the Kalman matrix. The reason for this is to avoid the discontinuity which occurs at 180“ from the nominal zero orientation.

When using quaternions to represent orientation and rotation in the Kalman filter, only three (x-y-z) of the four basis elements are used, because there are only three degrees of freedom in a unit quaternion (having too many degrees of freedom will 15 cause the state estimates to not converge). By storing quaternions directly (rather than using rotation angles), the frequent conversions between the two and the trigonometric functions this requires are avoided. At 100Hz, the angular changes are small enough that it makes negligible difference in the result, and if the discrete angular changes were larger, then, the linearity assumptions of the Kalman filter 20 break down anyway.

The filter update steps follow the typical sequence: * use an unscented transform to generate a representative set of state values * estimate the new state by propagating this set of values through the process function (incrementally rotating the orientation, and steady rotational velocity) 25 * calculate the mean and variance of the predicted state * add the process noise parameter * use an unscented transform to generate a representative set of predicted state values * estimate the measurements by propagating this set of values through the 30 measurement function (measured rotational velocity matching the rotational offeet, and measured accelerations matching a reoriented unit gravity) * calculate the mean and variance of the predicted measurements * add the measurement noise parameter (the acceleration noise variance is set to PCT/AU2015/000231 wo 2015/157808 8 10g, since it includes the contribution from player acceleration) * calculate the state/measurement covariance * calculate the Kalman gain (ratio of the state/measurement covariance and the measurement variance) 5 * calculate the new state mean and variance * accumulate the external orientation offset

The output of the Kalman filter is an absolute orientation. The facing of this orientation will drift over time. This is because without using magnetometers, the only inputs which affect facing, are the gyroscopes and gyroscopes measure 10 rotational velocity rather than orientation.

The program takes the absolute orientation from the filter and uses it to calculate a rotation that represents the tilt from vertical (ie, ignoring the facing component). This tilt is used to correct the accelerometer measurements, so that they are oriented with respect to the vertical axis, and the gravity component can then be subtracted 15 away.

Computing the Bowling Count

The bowling action is primarily characterized by a rotation of more than 500 7s around the upper back. This rotation has components in the coronal plane (around the forward axis) and in the transverse plane (around the up-down axis), with the 20 proportions depending on whether the bowling style is side-on or round-arm.

However, these are distinguished from throwing actions which are primarily in the transverse plane. A delivery must meet several preferred conditions in order to be counted. The rotational magnitude must be greater than 500 7s. The rotation in the coronal plane 25 must be at least two thirds of the rotation in the transverse plane. There must be a vertical acceleration peak of at least 4g in the 0.1 seconds prior to the rotation peak. And the run-up distance must be greater than 2.5 m.

Rotational Magnitude

The rotational magnitude is a measure of the intensity of the bowling action, which 30 may be useful for quantifying stresses on the body. It is calculated from the three gyroscope sensors, as the square root of the sum of squares of the rotational velocities. The peak value is reported. PCT/AU2015/000231 wo 2015/157808

Run-up distance

This distance is the total distance covered during the run-up prior to the delivery. This distance only begins when the velocity exceeds a predetermined value preferably 1.5 m/s, to exclude measuring the bowler walking to their mark. It is 5 calculated by integrating the GPS doppler velocity over the run-up interval.

Delivery Speed

The delivery speed in cricket depends on several factors: the forward motion of the bowler; the rotational velocity of their body in line with their arms; and to a lesser extent on shoulder rotation, elbow rotation, wrist movement, and the angle which the 10 ball leaves the hand.

This delivery speed can only be estimated, but both of the main factors may be measured. The forward velocity uses GPS doppler velocity, and the rotational contribution may be calculated from the rotational magnitude multiplied by the armspan of the bowler. The sum of these gives the estimated delivery speed. 15 In sports where a delivery involves a bent arm, the elbow still does not contribute significantly to the delivery speed. For example, in baseball, the slingshot action occurs at right angles to the elbow, with he energy of the action produced through forward motion, and rotation of the upper body. In these sports, an estimated delivery speed may be better calculated by a device mounted on the arm, preferably 20 near the forearm or wrist, rather than on the back.

Run-up velocity

The velocity at the point of delivery is taken from the GPS doppler velocity at the point of delivery. An alternative would be to average this velocity over the 5m or 10m prior to the delivery, which would bring it into line with traditional timing gate 25 methods of calculating this. 30

From the above it can be seen that this invention provides a unique means of collating training data and collecting performance data related to ball throwing. Those skilled in the art will realise that this invention may be implemented in other embodiment than those show without departing from the core teachings of this invention. For example this invention is applicable to other throwing and hitting actions, such as baseball deliveries or tennis serves.

Claims

1. A data collection system for measuring parameters of a human player in the act of throwing or serving a ball which includes a data collection device mounted on the upper torso of the player said device including a GPS unit, three dimensional accelerometers, three dimensional gyroscopes, and a microprocessor and data store for processing the signals from the GPS unit, three dimensional accelerometers and three dimensional gyroscopes said microprocessor being programmed to identify and collect signal data to derive one or more of delivery rotation speed, point of release and to derive delivery speed and maintain a count of throws or serves and optionally the distance and speed of the player during the act of delivery.

2. A data collection device as claimed in claim 1 adapted for cricket to derive bowling count, delivery speed, run up velocity and run up distance.

3. A data collection device as claimed in claim 2 wherein the run up distance measurement commences when the bowler’s velocity exceeds a predetermined value.

4. A data collection device as claimed in claim 2 wherein the run up distance is calculated by integrating the GPS doppler velocity over the run-up interval.

5. A data collection device as claimed in claim 2 wherein the run up velocity is taken from the GPS doppler velocity at the point of delivery.

6. A data collection device as claimed in claim 1 adapted for baseball pitching and tennis serving where the device is worn on the forearm or wrist.