Stairway to elders

The team provides complementary competences deployed in this project to achieve its main goal: designing and validating methods for continuous monitoring from multiple source of data of the emotional wellbeing and social interaction level of aging people, with strategic interventions from the environment to help them to get back both to positive emotions, in case of low-mood detection, and to a good level of social interaction.

• Aging population monitoring and Ambient Assistive Living

  Assistive environments and the potential benefit for aging population have been discussed in [Noceti2017; Martini2018] where we present a feasibility study for a shortstay post-hospitalization protected residence, built in the vicinity of a hospital, and designed as a smart environment with all the comforts of a private home.

• Motion analysis

  Motion information is the key element of the computation of a mobility index, we introduced in [Noceti2017], to evaluate the level of motility of elder people on the long period and in ecological conditions. The core functionalities of the proposed system are based on computer vision, and contribute to the assessment of the so-called frailty index, used by physicians to monitor the overall wellbeing of patients. Of relevance for this project are also previous work on motion understanding based on methods that capture human motion regularities. In [Noceti2017b], we introduced a computational model based on the Two-Thirds Power Law, and show how it can be used for visual action recognition; a multi-scale temporal extension of such model is discussed in [Vignolo2017] where we show its ability in characterizing the complexity and variability of a large class of movements. In [Malafronte2017] we investigated the use of space-time primitives to understand human movements and discuss the tolerance to view-point changes of the obtained representation.

• Face analysis and physiological signaling

  In [Cuculo2014] we present a facial landmark localization, demonstrating its effectiveness in presence of different expressions. In [D’Amelio2017] we design a model to pass from facial landmarks to the corresponding AU activations, exploiting the CANDIDE-3 face model. In [Boccignone2018] the processing of physiological data is discussed. After denoising, the wavelet transform is adopted to compensate the difference in activation time and persistence of each signal. Moreover, a in [Boccignone2017] a Multimodal dataset for HUmour SEnsing, AMHUSE, is presented along with a novel web-based annotation tool named DANTE (Dimensional ANnotation Tool for Emotions). The dataset collects the reactions of 36 subjects in presence of 3 amusing and 1 neutral video stimuli. Gathered data include RGB video and depth sequences along with physiological responses (electrodermal activity, blood volume pulse, temperature). The videos are annotated in terms of valence and arousal continuous dimensions.

• Emotion understanding

  Full-body movements description is at the basis of the system for emotion recognition we proposed in [Piana2016]. Information on kinematics and joints motion is used to extract expressive features inspired by psychology and humanistic theories, as e.g. contraction index, fluidity, and impulsiveness. In [Boccignone2018] we introduce a Bayesian framework based on deep latent representation of a continuous affective space which involves both facial cues and autonomic, physiological signals.

• Speed perception

  We have investigated speed perception of ordinary video clips in adults [de’Sperati2017; Rossi2018] and children [Zuliani2018], finding systematic and robust speed biases, which change significantly during childhood. We have put forth the hypothesis that one’s own motor rhythms can affect the sense of speed for external scenes, which in turn suggests that elders may prefer watching slower (or in some cases faster) TV programs.