Tracking von Bewegungen

Type: Semester thesis
Student: Daniel Höhener
Advisor: Corinne Mattmann
Project: Backmanager

It is necessary to be able to measure the postures and movements of the upper part of the body in order to prevent musculoskeletal disorders caused by work, sport or leisure time. The sensors used for this paper are completely integrable into clothing and hence permit a measurement which is comfortable for the user and independent of location. Up to now, postures could be recognized with strain sensors. These promising results were a motivation to test methods which recognize the movement of the upper part of the body.

This paper’s goal was to make a statement about the movements of the upper part of the body, with the help of strain sensors that were attached to a long-sleeved cat suit. For this purpose, two different tracking-methods were applied and tested. Beforehand, a reference system consisting of xSens movement sensors was attached to the prototype. This reference system is based on a model that splits up the upper part of the body into single segments. With that, the angles, and thus the degrees of freedom, could be defined for every segment.

In a next step, procedures of measurement were defined, which were the basis for the training and testing of the methods. An important component of the measurements was the recording of movements that explicitly concerned only single degrees of freedom. With those, the tracking-methods were trained and tested. For additional tests, measurements that contained movements with several combinations of joints were recorded.

The first implemented method of the degrees of freedom was based on the approach to assign a degree of freedom to a group of strain sensors. The main problem of this approach was the missing clearness, since a strain sensor can respond to more than just one degree of freedom. Neither with the second examined method of the neural networks could the problem of clearness be avoided completely. Due to this missing clearness, the results of the degrees of freedom are varying. The results of the two examined tracking-methods showed concordantly that the tested movements for the sequences of the degrees of freedom show smaller errors than for the sequences in which multiple joints have been combined. These errors are slight enough (median absolute error and standard deviation ~5°) to speak of a successful tracking. These results were mapped onto a matchstick man. One finding that could be drawn from the comparison of the two methods is that with the method of the degrees of freedom, the effort increases a lot faster than with the neural networks. It could be shown that movements, as long as they were tested separately and not as a combination of several degrees of freedom, can be tracked very well. With that, this paper informs about the possibilities but also about the challenges of tracking movements and serves as a basis for further works on strain sensors in textiles.