Jump detection

Algorithms to detect athletic jumps and to determine in real-time performance
parameters such as jump air time (AT), horizontal distance, height
and drop, are developed in this work. These algorithms are customized to be
implemented onboard action sports goggles developed by Recon Instruments
Ltd. These goggles are equipped with low cost micro-electro-mechanical inertial
sensors and a single point GPS receiver which feed raw data to the
algorithms. The micro-LCD display system in the goggles displays jump
statistics to the user wearing the goggles. Two novel methods, namely
WMCM (Windowed Mean Canceled Multiplication) and PFAD (Preceding
and Following Acceleration Dierence), are introduced for jump detection
using accelerometer data. Four characteristic points in the resultant
acceleration data are selected as the AT dening epochs for a jump. A
novel threshold independent, probabilistic method using MADM (Multiple
Attribute Decision Making) and the Closest Peak method are proposed to
detect these characteristic points and determine the corresponding AT of a
jump. A GPS/INS integration algorithm is developed to determine jump
horizontal distance, height and drop. A novel sensor error compensation
scheme is developed using sensor fusion and Linear Kalman Filters (LKF).
The LKF parameters are varied to address the
uctuating dynamics of the
athlete during a jump. The Extended Kalman Filter (EKF) used for GPS/
INS integration has an observation vector augmented with sensor error
measurements derived from sensor fusion.
The performance of the proposed algorithms was evaluated through experimental
eld tests. The proposed jump detection algorithm successfully
detected 92% of the jumps performed by a snowboarder wearing the goggles
whereas the current Recon algorithm only detects 60%. The AT determination
algorithm exhibited an average error of 0.033 s (4.8%) which is well
within the accuracy requirement of Recon, 0:1 s, and betters the current
Recon algorithm which has an average error of 0.111 s (8.4%). For determination
of jump horizontal distance, height and drop, the proposed algorithm
has an error of 14.34 cm (5.55%), 1.56 cm (38.21%) and 6.71 cm (9.43%)
respectively. The accuracy achieved is deemed to fulll expectations of both
recreational and professional athletes.