Bachelor's Thesis

Semi-supervised techniques for combined activity recognition and gait analysis


11/2018 – 04/2019


Home monitoring using wearables gives doctors access to continuous information about their patients allowing them to identify daily patterns of disease. For this to become a reality, accurate activity recognition and gait analysis systems are needed. Within the field of activity recognition there are multiple solutions available, however they rely on labelled training sets. This training data is time consuming and difficult to collect and label. One solution to this is the use of semi-supervised and unsupervised machine learning algorithms to decrease or possibly remove the need for labelled data. Two such algorithms which focus on the cyclic nature of walking are a local cyclicity estimator [3] and a iteratively trained hierarchical hidden Markov model [2]. Within the smart annotation field there are several solutions such as active labelling [4]. One can also approach the problem by using changes in entropy to identify the start of an activity [5] or unsupervised hierarchical clustering approaches [6]. Currently these solutions are implemented and tested in different contexts. This thesis aims to compare a selection of semi-supervised algorithms using a benchmark dataset [1] in order to understand the benefits and limitations of each approach in the context of combined activity recognition and gait analysis for home monitoring applications. The most successful of these algorithms will then be used to add labels to existing public databases in order to increase the data accessible for training.


  1. Martindale, C., Roth, N., Hannink, J., Sprager, S., & Eskofier, B. (2018). Smart Annotation Tool for Multi-sensor gait based daily activity data. In 2018 IEEE International Conference on Pervasive Computing and Communications Workshops (PerCom Workshops). Athens.
  2. Martindale, C., Hoenig, F., Strohrmann, C., & Eskofier, B. (2017). Smart Annotation of Cyclic Data Using Hierarchical Hidden Markov Models. Sensors, 17(10), 2328.
  3. Sprager, S., & Juric, M. (2018). Robust Stride Segmentation of Inertial Signals Based on Local Cyclicity Estimation. Sensors, 18(4), 1091.
  4. Diete, A., Sztyler, T., Stuckenschmidt, H., Diete, A., Sztyler, T., & Stuckenschmidt, H. (2018). Exploring Semi-Supervised Methods for Labeling Support in Multimodal Datasets. Sensors, 18(8), 2639.
  5. Sadri, A., Ren, Y., & Salim, F. (2016). Information gain-based metric for recognizing transitions in human activities. Pervasive and Mobile Computing, 38(1), 92–109.
  6. Feng Zhou, De la Torre, F., & Hodgins, J. K. (2013). Hierarchical Aligned Cluster Analysis for Temporal Clustering of Human Motion. IEEE Transactions on Pattern Analysis and Machine Intelligence, 35(3), 582–596.