A novel modeling approach to fall detection and experimental validation using motion capture system

The injuries caused by falls are great threats to the elderly. As a consequence, fall detection has been considered to play an important role in monitoring the security and well-being of the elderly who live alone. However, many fall detection systems suffer from the missed detection and false alarm. This paper presents a novel modeling approach to fall detection using data from motion capture system. In the proposed model, angle characteristic which appears simultaneously with the maximum of acceleration is combined with acceleration characteristic together for distinguishing falls from activities of daily living (ADL). Five features are extracted from each activity as a sample and a SVM (Support Vector Machine) classifier is gained by training the sample set. The activity types of unknown samples are predicted using the classifier gained above. The proposed model can achieve convincing fall detection results with different subjects while maintaining a high sensitivity of 100% and specificity of 94%. The effectiveness of the proposed approach has been validated by experiments.