This chapter presents a hardware/software platform based on a state-of-the-art Time-of-Flight (ToF) sensor and a low-power embedded computing system for the automated recognition of body postures with applications ranging from detection of dangerous events (e.g. falls) to natural human-computer interaction (e.g. assistance during rehabilitation/training exercises). The platform meets typical requirements for Ambient Assisted Living (AAL) applications such as compactness, low-power consumption, noiseless, installation simplicity, etc. In order to accommodate several application scenarios, satisfying different requirements in terms of discrimination capabilities and processing speed, two feature extraction approaches are investigated (namely topological and volumetric) and related performances are compared. Discrimination capabilities of the two approaches are evaluated in a supervised context, achieving a classification rate greater than 96.5 %. The two approaches exhibit complementary characteristics achieving high reliability in several scenarios in which posture recognition is a fundamental function. © 2014 Springer Science+Business Media.
Time-of-flight sensor-based platform for posture recognition in AAL applications
Leone Alessandro;Diraco Giovanni;Siciliano Pietro
2014
Abstract
This chapter presents a hardware/software platform based on a state-of-the-art Time-of-Flight (ToF) sensor and a low-power embedded computing system for the automated recognition of body postures with applications ranging from detection of dangerous events (e.g. falls) to natural human-computer interaction (e.g. assistance during rehabilitation/training exercises). The platform meets typical requirements for Ambient Assisted Living (AAL) applications such as compactness, low-power consumption, noiseless, installation simplicity, etc. In order to accommodate several application scenarios, satisfying different requirements in terms of discrimination capabilities and processing speed, two feature extraction approaches are investigated (namely topological and volumetric) and related performances are compared. Discrimination capabilities of the two approaches are evaluated in a supervised context, achieving a classification rate greater than 96.5 %. The two approaches exhibit complementary characteristics achieving high reliability in several scenarios in which posture recognition is a fundamental function. © 2014 Springer Science+Business Media.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.