In past years, high-intensity event analysis in sports applications, e.g. soccer, tennis, golf
or basketball has been the focus of many researchers. Analyzing the shot in soccer, the serve
in tennis, the stroke in golf or the lay-up in basketball provides important feedback for athletes
to improve their performance. Therefore, an individual monitoring of athletes during training
sessions or game situations is mandatory. Video-based systems are commonly used for monitoring
but suffer from several disadvantages. These systems are expensive, time-consuming to set up,
and are mainly applicable in a lab environment under controlled conditions, e.g. movements in
a limited capture volume. In recent years, small, lightweight, and inexpensive inertial sensors
provide an objective and unobtrusive method to acquire motion data under realistic conditions,
e.g. on the pitch or court. Due to highly dynamic movements in realistic conditions, multiple
sensors might be used to increase the performance of event analysis systems.
Therefore, the goal of this master thesis is the development of sensor fusion techniques for highintensity
event analysis in sports and the evaluation under realistic conditions, e.g. on the pitch.The following tasks have to be conducted:
Data Collection
The student has to define one high-intensity event in one particular sports application, e.g. a lay-up in basketball. Due to highly dynamic motions, multiple sensors on different body
positions, e.g. hip, chest, and ankle, have to be used for data acquisition. The student has
to define one parameter that can be estimated for the defined event, e.g. the ball speed.
A video-based system should be used as ground truth for the evaluation. The student has
to setup a study protocol, recruit at least ten subjects and perform the study. The study
protocol should consider movements under realistic conditions, e.g. an exercise consisting
of dribbling, pass, and lay-up.
Preprocessing
The student has to develop preprocessing techniques including e.g. low-pass filtering, segmentation,
and extraction of biomechanical features [1, 2].
Sensor Fusion Technique for Parameter Estimation
The student has to develop an algorithm which estimates the self-defined parameter. Since
multiple sensors are acquired, sensor fusion techniques have to be considered [3].
Performance Assessment
In the performance assessment, the estimated parameter has to be compared to the videobased
ground truth system. Using multiple sensors on different body positions requires the
investigation of different sensor position configurations.
For this thesis, a profound research on literature, patents and related work in the corresponding
working areas has to be performed. Current software standards like maintainability,
reusability and appropriate documentation have to be maintained. The master thesis is supported by the adidas AG.
References
[1] A. Bulling, U. Blanke, and B. Schiele "A Tutorial on Human Activity Recognition Using
Body-Worn Inertial Sensors" ACM Comput Serv, 46(3):33:1-33:33, 2014.
[2] J. Favre, B. M. Jolles, R. Aissaoui, and K. Aminian "Ambulatory measurement of 3D knee
joint angle" J Biomech, 41(5):1029-1035, 2008.
[3] D. Hall, and J. Llinas, "An Introduction to Multisensor Data Fusion" Proc. IEEE, 85(1):6-
23, 1997.