Skip to content
BY-NC-ND 4.0 license Open Access Published by De Gruyter September 7, 2017

Development of a retrospective process for analyzing results of a HMM based posture recognition system in a functionalized nursing bed

  • Julia Demmer EMAIL logo , Andreas Kitzig and Edwin Naroska


In the area of care in general but especially in the care of elderly, there is a great interest in deriving patient parameters preparation free. For this purpose, a load cell functionalized nursing bed has been developed at Niederrhein University of Applied Science. The system allows analysis and recognition of the persons’ positions and actions in the bed. The Hidden Markov Toolkit (HTK) based posture recognition system was initially presented at the BMT 2015 by our research group. The initial system shows good results but to draw conclusions about the patient's condition, a minimum possible error rate should be achieved. For this purpose, a two-step retrospective analysis of the initial results was developed as an extension to improve the accuracy of the system.

In a first step, the results of the initial recognition are analysed and classified into correct or incorrect results. This is done by means of probability based distance measure. The distances are calculated by the Viterbi algorithm. Based on these distance measures, errors and so called ‘secure support points’ are determined in the recognition result and the associated parameters are extracted.

The second part of the extension deals with a retrospective recognition method to improve the error rate of the initial recognition step. Based on the determined support points, the errors are analysed again applying syntactic relation. To this end, a special parameter set to control the recognition system, the so-called ‘syntax’, is generated individually for each erroneous recognition result based on the preceding secure support point. The new parameterized recognition system is then used to improve the initial erroneous results by re-classification. Finally, the corrected recognition results are combined with the preliminary recognition result.

Published Online: 2017-09-07

©2017 Julia Demmer et al., published by De Gruyter

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Downloaded on 8.12.2023 from
Scroll to top button