Reference: A Wixted, W Spratford, M Davis, M Portus, D James, Wearable Sensors for On FIeld Near Real Time Detection of Illegal Bowling Actions, 1 of 1-Conference of Science, Medicine & Coaching in Cricket 2010, 165
A well-developed and legal bowling action is of enormous importance to developing
cricket athletes and teams both nationally and internationally. Good bowlers can make the
headlines, but so do the controversial ones. The bowling action can currently only be measured
with accuracy in a few specialist laboratories around the world. Testing requires a break in a
player’s schedule, travel to a distant testing centre and players are then asked to bowl in an
unnatural environment. This testing is therefore somewhat expensive and not routinely accessible,
but this process is the best method currently available.
Recently developed micro technologies have been used in athlete performance monitoring,
biomechanical monitoring and physiological monitoring in other sports such as swimming
(Kavanagh et.al. 2006, James et.al. 2007, Davey et.al. 2008), snowboarding (Harding 2008),
running (Wixted et.al 2007, 2010) and also in cricket (Rowlands et.al. 2009). These technologies
offer the promise of wearable alternatives to laboratory testing that would be inexpensive enough
to be used routinely by elite and developmental players and coaches. In light of the above, real
time, non-laboratory monitoring of the bowling arm-action by miniature sensors is a potential
solution to the disruptive issues caused by the processes currently in-place to assess perceived or
actual throwing. Current and future miniature sensors are achieving, or will soon achieve, the
sensitivity and range of operation necessary to monitor athletes operating at the elite levels of
many sports. In the case of monitoring the arm action of bowlers, a number of checkpoints needed
to be achieved to verify that miniature sensors could supply the required level of monitoring. The
verification process involved testing with actual sensors and the use of “virtual” sensors generated
from the existing library of Cricket Australia motion capture data (Vicon). These combined tests
identified some of the sensor operational ranges and the feasibility of identifying the illegal arm
actions using miniature sensors.
Sensor systems have a very restricted view of the world and this limits their ability to identify
activities that seem obvious to an onlooker. Sport involves movement and the sensors
predominately used for measuring kinematic activities are accelerometers and rate gyroscopes (to
measure rotation). These sensors are often arranged on three orthogonal axes to capture activity
in three dimensions (3D). The accelerometers and rate-gyroscopes are usually Micro-Electro
Mechanical Systems (MEMS) and are manufactured, along with support electronics, directly in
silicon with sizes of 3D accelerometers as small as 5x3x1mm. Gyroscopes have similar sizes. The
combination of 3D accelerometers and gyroscopes and the fast sampling rate of sensor systems
allow the sensor systems to capture information about a small segment of the physical world yet
this small segment is often sufficient to determine the parameters of interest.
To monitor a bowler for illegal bowling it is necessary that the sensor can identify specific points of
the arm action, in particular the start and end of the arm action. The start of the arm action occurs
when the elbow is level with the shoulder, the end of the arm action occurs when the ball is
released. Between these two points the sensor needs to monitor the critical aspect of elbow
extension. Although stationary 3D accelerometers can detect orientation, the complex forces
affecting a moving limb prevent the direct extraction of orientation therefore; a simple orientation
based detection of start of arm action is not possible. The ball release point appears to occur
around 120 degrees after the start of arm action. Other than some sort of finger tip pressure sensor
there does not appear to be any obvious sensor measure that can detect ball release. Finally, the
detection of elbow extension: this does appear amenable to detection by miniature sensors. Using
data from multiple accelerometers or multiple gyroscopes, elbow extension should be detectable
as a centrifugal acceleration phase difference or a rotation rate phase difference between upper
arm and forearm mounted sensors.
To ascertain the feasibility of MEMS based sensor monitoring of bowling arm elbow extension, the
dynamics of the bowling arm were investigated using ‘virtual’ sensors. The virtual sensors were
also used to determine the working range required for physical sensors.
Virtual Sensors: 120 Hz Vicon motion capture data from numerous bowlers performing a range of
deliveries was processed to generate the output of virtual sensors. This process involved selecting
specific Vicon markers and converting their position data from an external ‘cricket-pitch’ based
Frame of Reference (FOR) to acceleration data and angular velocity data in a sensor based FOR.
The markers involved were part of a semi rigid technical marker set. The process of creating the
virtual sensors involved double differentiation of the position information to convert it to
acceleration. For each frame of Vicon data the orientation of the technical marker relative to the
external FOR was extracted and the orientation used to transform the external FOR acceleration
into sensor FOR acceleration. As the process of transforming acceleration from one FOR to
another involved generating rotation angles, the sequence of rotation angles was differentiated to
give rotation rates (angular velocity).
The data from the virtual sensors represented the outputs of accelerometers and rate-gyroscopes
had they been fitted in the location of the technical marker when the original motion capture data
was collected. This data was analysed and a very clear ball-release signature was identified
indicating that real sensors should be able to identify the ball release point (Fig.1(a)). Not included
in Fig.1(a) were two illegal bowls where there was a distinctive phase difference between the
upper-arm and forearm peak acceleration.
No specific start of arm-action was initially identified in 120Hz Vicon data but when data from a 250
Hz Vicon system was processed many deliveries showed a spike of very high angular velocity on
or about the start of the bowling action (Fig.1(b)). Video replay indicated that this was associated
with the elbow extension and shoulder rotation that brings the ball into the upward facing direction
(a supination action) just prior to the start of the arm action. This shoulder rotation is necessary to
prepare the shoulder joint for the delivery action.
Figure 1: (a) Centrifugal Acceleration for right arm bowlers at ball release, (b) Example of two axes
of angular velocity with a spike of angular velocity at the start of the arm action.
Despite the apparent confirmation of the existence of potential arm-action start and end signatures,
the Vicon data is subject to its own processing errors and these signatures need to be confirmed
with actual sensors. The Vicon data does give indicative acceleration and angular velocity:
• Forearm Centrifugal Acceleration: > 70g ( ~ 700ms-2)
• Forearm Angular Velocity: > 3500 deg/s
Both these values are achievable by current technology.
The positions of the Vicon technical markers were not the optimal positions for
actual sensors attempting to detect elbow extension. In a separate experiment rate-gyroscopes
were used to synchronously collect data from either side of the elbow joint. Elbow extension and
flexion was performed through a bowling action. From the gyroscopes, an arm starting in the
already flexed position and extending through the bowl showed peak rotation rates for the forearm
occurring prior to the upper-arm peak rotation (figure 2). An arm staying straight through the
bowling action and then flexing after release showed the gyroscopes for the upper arm and
forearm tracking together until the flexing started (figure 3). The gyroscope output for an arm
starting straight but flexing through the action showed the sensors not even tracking together with
the forearm’s sensors peaking sometime after the upper arm sensor (figure 4)
Figure 2: Exaggerated stick figure representation and gyroscope output for an arm starting the bowling
action flexed and extending through to release.
Figure 3: Exaggerated stick figure representation and gyroscope output for an arm starting the
bowling action straight and beginning to flex at release.
Figure 4: Exaggerated stick figure representation and gyroscope output for an arm starting the
bowling action straight and flexing through the bowl.
Where the arm remains straight through the bowling action it appeared that the outputs of the
sensors tracked together. If a single flexion or extension occurred the phase relationship between
the forearm and upper arm sensors was distinctly different. At this time the effect of a complete
flexion-extension action during the bowling has not been analysed although study of the historical
data indicates that this type of action occurs for some bowlers.
Similar data has been collected from accelerometers and although the accelerometers used for
these experiments were overloaded by the bowling action, the phase shifting between upper-arm
sensors and forearm sensors for flexion-extension was identified in the output.
Analysis from virtual sensors indicated that arm action start and end points have
identifiable signatures that can be utilised by miniature sensors. The virtual sensors also indicated
the signal magnitudes. For some types of illegal bowling action, the virtual accelerometers
identified a phase shift between forearm and upper-arm acceleration. Analysis of the output of
physical sensors indicated that changing elbow extension during the bowling arm action is
identifiable in the sensor output. These factors have become the input to the development of
sensors specifically designed to capture the bowling action and extract elbow extension. The next
phase of this development is the confirmation of sensor operation using simultaneously collection
of data from both the sensors and the motion capture system. This step will also begin the
development of the mapping from sensor phase shift to elbow angle
Acknowledgements: The authors acknowledge the funding for this project from the ICC and the
Marylebone Cricket Club (MCC) as well as the support and historical data from CA and the AIS
Kavanagh JJ, Morrison S, James DA, Barrett R, Reliability of segmental accelerations measured
using a new wireless gait analysis system, Journal of Biomechanics, Vol 39, pp2863-72, 2006
James DA, Davey N, Hayes J, From conception to reality: A wearable device for automated
swimmer performance analysis, Japan Society for Sciences in Swimming and Water Exercise 11th
Annual Conference proceedings, pp101-109. 2007
Davey N, Anderson M, James DA, Validation trial of an accelerometer-based sensor platform for
swimming, Sports Technology, 2008; 1 (4) pp 202-207
Harding JW, Macintosh CG, Martin DT, James DA, Classification of aerial acrobatics in elite halfpipe
snowboarding using body mounted inertial sensors, The Engineering of Sport, 2008, Estivalet,
Brisson Ed. pp447-456 Paris: Springer
Wixted AJ, Thiel DV, Hahn A, Gore C, Pyne D, James DA, Measurement of Energy Expenditure in
Elite Athletes using MEMS based inertial sensors, IEEE Sensors Journal, Vol 7, No4, pp481-8,
Wixted AJ, Billing DC, James DA, Validation of trunk mounted inertial sensors for analysing
running biomechanics under field conditions, using synchronously collected foot contact data.
Sports Engineering, 2010, In Press.
Rowlands D, James DA, Thiel DV, Bowler Analysis in Cricket using Centre of Mass Inertial
Monitoring, Sports Technology, 2009, (21-2), pp39-42.