Development of a novel algorithm for human fall detection using wearable sensors Gaetano Anania, Alessandro Tognetti, Nicola Carbonaro, Mario Tesconi, Fabrizio Cutolo, Giuseppe Zupone, Danilo De Rossi Interdepartmental Reserch Center “E.Piaggio” University of Pisa Pisa, Italy Email: g.anania@ing.unipi.it Abstract— A novel algorithm for human fall detection by means of a tri-axial accelerometer, is described. A module constituted by the accelerometer and an on board processing unit was designed and realized. The system is conceived to be used in a multi-sensor network context for the remote monitoring of personnel working in very severe conditions (firefighters and civil protection operators). In the real application the module is thought to be integrated in the operator uniform collar. The algorithm is based on the detection of a critical trunk inclination in correspondence of an high rotational velocity. A Kalman filter was designed in order to separate the signal component due to gravity (i.e useful to extract the subject orientation) from the one due to the system acceleration. In comparison with the existing solutions the realized algorithm presents many advantages: no training is needed, low computational costs, fast time response and good performances also during critical activities (e.g jumping, running). I. I NTRODUCTION The recent developments in sensor miniaturization and wireless communication technologies have enabled the devel- opment of affordable wearable monitoring systems. Building portable devices containing sensors and electronics for signal processing and wireless data transmission has become easier and cheaper. The main advantage of these systems is that they can be worn by subjects also when performing their daily ac- tivities. In the last few years, the use of portable devices in the biomedical field has considerably increased, especially in the health monitoring of chronic patients. Moreover, another very challenging area is the use of wearable technologies for the monitoring of physiological and environmental parameters of people performing their working activities. In particular, in the case of operators dealing with dangerous conditions (as Fire- Fighters or Civil Protection’s rescuers), a continuous real-time monitoring is very important to prevent injuries and dangerous situations [1]. The aim of this work was the development of a novel algorithm for human fall detection. This work has been carried out within the ProeTEX project (FP6 IST4026987) whose aim is to develop a multi-sensor wearable system for the support of emergency personnel (firefighter and civil pro- tection operators). ProeTEX is developing innovative garments for the market of emergency operators. These garments contain a network of sensors that keeps permanently under control the health state of the wearer [2]. According to the project users’ requirements one of the most important parameter to be monitored is the operator fall to the ground which can be related to a very dangerous situation for the operator. Considering the target application, the portable system must be active for several hours without re-charging the batteries; hence the use of low power electronics with limited memory and processing resource is mandatory. Moreover, to reduce the amount of transmitted data a real time on board processing is necessary. For these reasons the main requirements for the developed algorithm were: (1) On line data processing; (2) low computational costs and possibility to be implemented on a low power and limited memory platform; (3) high reliability also during “rapid” activities such as running or jumping which are typical of emergency operators. In the realized prototype a tri-axial accelerometer and the related electronics was integrated in an operator jacket on the upper part of the trunk. The commercial MEMS accelerometers have low sizes and low power consumption and they can be easily integrated in portable units. Tri-axial accelerometers are widely used in order to detect posture or classify human ac- tivities for ambulatory monitoring in the field of rehabilitation and elderly surveillance [3], [4]. In the scientific literature several works dealing with algorithms for fall detection with accelerometers exist [4]–[9] with a particular focus on elderly monitoring. However, these systems do not completely fulfill the needed requirements. The algorithms presented in [7] and [9] use pattern recognition techniques thus needing an extensive training phase. Considering the application, the fall detection device has to be used “as it is” without training efforts for the user. In [4]–[6] three different fall detection systems are presented in which the accelerometer is worn in different locations (waist, trunk and thigh); the detection algorithms apply certain thresholds on the signal magnitude (SM, defined as the root sum of squares of the accelerometer components) in order to identify the peculiar parts of a fall event: the free fall (SM goes below a certain threshold th l ) and the impact (SM exceeds a threshold th h , with th h > th l ); these algorithms show good sensitivity (greater than 90%) and specificity tested on a wide range of activities of daily living. For the monitoring of emergency workers, different problems have to be considered respect to the elderly surveillance case. 1-4244-2581-5/08/$20.00 ©2008 IEEE 1336 IEEE SENSORS 2008 Conference