Proposal of a Similarity-based Unhealthy Motion Detection Method for the Elderly using DTW and MT Method

There is a demand for quantitative evaluation of the motor function of the elderly in the preventive care activities or the home visit rehabilitation. RGB-D sensors are promising devices for the application due to its portability and data acquisition ability. This paper proposed an unhealthy motion detection method that obtained skeleton data from a measurement system using the RGB-D sensor, extracted motion feature values from the time series data by the DTW method, and detected unhealthiness of the motions from the values using the MT method based on similarities to motions of able-bodied persons. The method was applied to existing motion data. The unit space of MT method was constructed from 30 able-bodied persons, and the judgements on 7 types of motion for the 23 elderlies were made. The comprehensive indices for each type of motion for each elderly were defined by integrating visual judgment of 4 physical therapists, and concordance rates of the judgement were investigated. It was observed that for 3 types of motion from 7 the proposed method shows equal to or better than lowest score of the physical therapist. It was confirmed that the rates of contribution for the feature values of the motions judged as unhealthy shows tendency in relation to a wobbling motion or lack of joint range.