Parkinson's disease (PD) is a common neurodegenerative disorder characterized by disabling motor and non-motor symptoms. For example, idiopathic hyposmia (IH), which is a reduced olfactory sensitivity, is typical in >95% of PD patients and is a preclinical marker for the pathology. In this work, a wearable inertial device, named SensHand V1, was used to acquire motion data from the upper limbs during the performance of six tasks selected by MDS-UPDRS III. Three groups of people were enrolled, including 30 healthy subjects, 30 IH people, and 30 PD patients. Forty-eight parameters per side were computed by spatiotemporal and frequency data analysis. A feature array was selected as the most significant to discriminate among the different classes both in two-group and three-group classification. Multiple analyses were performed comparing three supervised learning algorithms, Support Vector Machine (SVM), Random Forest (RF), and Naïve Bayes, on three different datasets. Excellent results were obtained for healthy vs. patients classification (F-Measure 0.95 for RF and 0.97 for SVM), and good results were achieved by including subjects with hyposmia as a separate group (0.79 accuracy, 0.80 precision with RF) within a three-group classification. Overall, RF classifiers were the best approach for this application. The system is suitable to support an objective PD diagnosis. Further, combining motion analysis with a validated olfactory screening test, a two-step non-invasive, low-cost procedure can be defined to appropriately analyze people at risk for PD development, helping clinicians to identify also subtle changes in motor performance that characterize PD onset.

Recognition of activities of daily living plays an important role in monitoring elderly people and helping caregivers in controlling and detecting changes in daily behaviors. Thanks to the miniaturization and low cost of Microelectromechanical systems (MEMs), in particular of Inertial Measurement Units, in recent years body-worn activity recognition has gained popularity. In this context, the proposed work aims to recognize nine different gestures involved in daily activities using hand and wrist wearable sensors. Additionally, the analysis was carried out also considering different combinations of wearable sensors, in order to find the best combination in terms of unobtrusiveness and recognition accuracy. In order to achieve the proposed goals, an extensive experimentation was performed in a realistic environment. Twenty users were asked to perform the selected gestures and then the data were off-line analyzed to extract significant features. In order to corroborate the analysis, the classification problem was treated using two different and commonly used supervised machine learning techniques, namely Decision Tree and Support Vector Machine, analyzing both personal model and Leave-One-Subject-Out cross validation. The results obtained from this analysis show that the proposed system is able to recognize the proposed gestures with an accuracy of 89.01% in the Leave-One-Subject-Out cross validation and are therefore promising for further investigation in real life scenarios.

Robots will become part of our everyday life as helpers and companions, sharing the environment with us. Thus, robots should become social and able to naturally interact with the users. Recognizing human activities and behaviors will enhance the capabilities of the robot to plan an appropriate action and tailor the approach according to what the user is doing. Therefore, this paper addresses the problem of providing mobile robots with the ability to recognize common daily activities. The fusion of heterogeneous data gathered by multiple sensing strategies, namely wearable inertial sensors, depth camera, and location features, is proposed to improve the recognition of human activity. In particular, the proposed work aims to recognize 10 activities using data from a depth camera mounted on a mobile robot able to self-localize in the environment and from customized sensors worn on the hand. Twenty users were asked to perform the selected activities in two different relative positions between them and the robot while the robot was moving. The analysis was carried out considering different combinations of sensors to evaluate how the fusion of the different technologies improves the recognition abilities. The results show an improvement of 13% in the F-measure when different sensors are considered with respect to the use of the sensors of the robot. In particular, the system is able to recognize not only the performed activity, but also the relative position, enhancing the robot capabilities to interact with the users.

Activity Recognition is important in assisted living applications to monitor people at home. Over the past, inertial sensors have been used to recognize different activities, spanning from physical activities to eating ones. Over the last years, supervised methods have been widely used, but they require an extensive labeled data set to train the algorithms and this may represent a limitation of concrete approaches. This paper presents a comparison of unsupervised and supervised methods in recognizing nine gestures by means of two inertial sensors placed on the index finger and on the wrist. Three supervised classification techniques, namely random forest, support vector machine, and multilayer perceptron, as well as three unsupervised classification techniques, namely k-Means, hierarchical clustering, and self-organized maps, were compared with the recognition of gestures made by 20 subjects. The obtained results show that the support vector machine classifier provided the best performances (0.94 accuracy) compared with the other supervised algorithms. However, the outcomes show that even in an unsupervised context, the system is able to recognize the gestures with an average accuracy of ~0.81. The proposed system may be therefore involved in future telecare services that could monitor the activities of daily living, allowing an unsupervised approach that does not require labelled data.

Daily activity recognition can help people to maintain a healthy lifestyle and robot to better interact with users. Robots could therefore use the information coming from the activities performed by users to give them some custom hints to improve lifestyle and daily routine. The pervasiveness of smart things together with advances in cloud robotics can help the robot to perceive and collect more information about the users and the environment. In particular thanks to the miniaturization and low cost of Inertial Measurement Units, in the last years, body-worn activity recognition has gained popularity. In this work, we investigated the performances with an unsupervised approach to recognize eight different gestures performed in daily living wearing a system composed of two inertial sensors placed on the hand and on the wrist. In this context our aim is to evaluate whether the system is able to recognize the gestures in more realistic applications, where is not possible to have a training set. The classification problem was analyzed using two unsupervised approaches (K-Mean and Gaussian Mixture Model), with an intra-subject and an inter-subject analysis, and two supervised approaches (Support Vector Machine and Random Forest), with a 10-fold cross validation analysis and with a Leave-One-Subject-Out analysis to compare the results. The outcomes show that even in an unsupervised context the system is able to recognize the gestures with an averaged accuracy of 0.917 in the K-Mean inter-subject approach and 0.796 in the Gaussian Mixture Model inter-subject one.