Optimization and evaluation of motion sequences of an averaged human motion model by using kinematic techniques and new evaluation methods

Introduction Biosignal processing, pattern recognition, modelling and simulation require a large amount of reference data, both for the development of new algorithms and for evaluation. Depending on the application, the availability of databases is rather low. In the field of biosignal processing with a focus on the functionalization of (nursing) beds, a database with motion sequences of persons in a nursing bed and a method to create average human motion sequences e.g., for person-independent simulation experiments was presented earlier by our research group. Evaluations revealed that some of the averaged motion patterns contain interfering artifacts that arise after averaging due to the variability of the original motion patterns, and thus are not directly applicable. Therefore, we created additional methods for optimizing, evaluating, and merging averaged sequences. Methods An additional procedure was created that realizes the optimization and merging of averaged motion sequences, but also the suppression of unsuitable motion artifacts, with considerations from the field of kinematics. This is done by a processing chain in which, in addition to kinematic considerations, observations of the process-oriented behaviour of human movements, physiological considerations and appropriate interpolation and filtering methods are incorporated. Before and after the merging and post-processing steps, the motion sequences must be tested and evaluated for use in simulation tasks. For this purpose, four general evaluation criteria are proposed (temporal change of the body vector, the roughness of the motion sequence, the speed of change of a motion sequence and a measure of volume utilization), which enable an assessment of motion sequences in addition to a visual evaluation. Results & Conclusion The presented method for combination and optimization was subjected to an evaluation against the four presented criteria. It has been shown that combining and optimizing motion sequences does not result in sequences that are fully comparable to real human motion sequences, but the combined sequences are close to real sequences and can be used for simulation tasks. The suitability of the evaluation criteria was evaluated and confirmed using visual comparisons. The new criteria can contribute to a more objective evaluation of motion sequences. The approach will be further improved in the future, for example by analyzing the spectral composition of the individual angular progressions to detect and suppress disturbing artifacts in an even more precise manner.