Physiotherapy Exercises Recognition Based on RGB-D Human Skeleton Models

In the Western world, aging is a growing problem of the society and computer assisted treatments can facilitate the telemedicine for old people or it can help in rehabilitations of patients after sport accidents in far locations. Physical exercises play an important role in physiotherapy and RGB-D devices can be utilized to recognize them in order to make interactive computer healthcare applications in the future. A practical model definition is introduced in this paper to recognize different exercises with Asus Xtion camera. One of the contributions is the extendable recognition models to detect other human activities with noisy sensors, but avoiding heavy data collection. The experiments show satisfactory detection performance without any false positives which is unique in the field to the best of the author knowledge. The computational costs are negligible thus the developed models can be suitable for embedded systems.