Authentication of persons using acceleration sensors in a telemedicine setting

Prophylactic measures are reducing costs of healthcare without lowering its quality. Telemetric Personal Health Monitoring (TPHM) Systems with multi-parametric sensors allow active integration of patients into therapy. A common problem in such systems is the verification and identification of users to guarantee the authenticity of gathered data. To assign measured physiological parameters to legitimate users we included a gait-based identification algorithm into a TPHM System. The algorithm records two-dimensional acceleration data from a two-axis accelerometer located on a chest belt. In configuration mode the participant has to walk fast to allow the algorithm to calculate a reference pattern for the legitimate user of the belt. In identification mode the algorithm is looking for fast walking sequences and compares the recorded pattern with the reference pattern. Test patterns generated by different users wearing the belt are compared to find out if the algorithm is able to distinguish between legitimate and illegitimate users. We use methods similar to classification procedures in speech processing to analyze our data in the frequency domain. Several approaches and classificators like the Mahalanobis Classificator and the Minimum Distance Classificator were examined. The signal processing chain consisted of Hamming Windowing, FFT (Fast Fourier Transformation), calculation of the performance density spectrum and calculation of Fourier coefficients . Furthermore, histogram statistics were used as a simple and performance-efficient way to classify the acquired data. We showed that it is difficult to use acceleration data to identify persons without large training sets.