Système de télésurveillance sonore pour la détection de situations de détresse

The telemedicine and medical remote monitoring in particular, today represents an effective solution to the health professional shortcomings facing to the increasing older population. In addition to the comfort of being at home, this system decreases the cost of long hospitalization. The proposed system achieves the home medical telesurveillance by means of microphones and medical sensors to detect a distress situation. This system is designed for elderly people at home or for the patient with chronic illness. The experimental apartment is equipped with medical sensors (tensiometer, oxymeter, balance, etc.), infrared position sensors, and acoustic sensors (1 microphone/room). The originality of the system comes from the replacement of the video surveillance with a sound surveillance. The sound analysis system is divided in two stages: firstly, the detection/extraction of the sounds is operated and secondly, a classification of these sounds in known classes takes place. A wavelet-based algorithm with good performance when applied in noisy environments is proposed. The acoustical classification step uses a Gaussian Mixture Models to classify the sounds according to the 7 predefined classes. The detection algorithm allows an equal error rate of 0% for the signal to noise ratio superior or equal to 10 dB and 4% for the 0 dB. The proposed system coupled with a classical medical telesurveillance system will bring extra information needed for the reliability of the global system.