An Experimental Study on New Features for Activity of Daily Living Recognition Daniele Ferretti, Emanuele Principi, Stefano Squartini Department of Information Engineering Universit` a Politecnica delle Marche Via Brecce Bianche, 60131, Ancona, Italy d.ferretti@pm.univpm.it, {e.principi,s.squartini}@univpm.it Luigi Mandolini MAC Srl Via XX Settembre 21, 62019 Recanati (MC), Italy mandolini@mac-italia.com Abstract—In the last few years, the researchers have spent many efforts in developing advanced systems for activity daily living (ADL) recognition in diverse applicative contexts, as home automation and ambient assisted living. Some of these need to know in real time the actions performed by a user, and this involves a number of additional issues to be taken into account during the recognition. In this paper, we present some improvements of a sliding window based approach to perform ADL recognition in a online fashion, i.e., recognizing activities as and when new sensor events are recorded. We describe seven methods used to extract features from the sequence of sensor events. The first four relate to previous works regarding the system of ADL recognition described, while, the last three represent the original contribution of this work. Support Vector Machine (SVM) has been used as classifier. Several experiments have been carried out by using a public smart home dataset and obtained results show that two of the three novel approaches allow to improve the recognition performance of the conventional methods, up to an increment of 5% with respect to the baseline feature extraction approach. I. I NTRODUCTION The recent advances in the field of pervasive and ubiquitous computing have made possible the development of many applications that base their operation on the recognition of activity of daily living (ADL). This task will be named as activity recognition (AR) in the following, for the sake of brevity. Some examples of these applications are the health care systems for the elderly and disabled people [1], [2], context-aware prompting systems [3], [4], surveillance systems [5] and interactive gaming interfaces [6]. These systems need to recognize the activities that a person is carrying out in a non invasive manner, by observing the behaviour of people and the environment from sensor readings. In this research area, several studies exist that deal with the problem using different methods and approaches. These works can be classified on the basis of their characteristics. A first classification criterion is to distinguish between knowledge- driven and data-driven approaches. Data-driven techniques use the information provided by the sensors to calibrate or build models of human activities in a supervised or unsupervised manner [7], [8]. Differently, knowledge-driven techniques ap- proach to the AR task from a different point of view, i.e., by creating a priori models of the activities based on the knowledge of the problem, according to a logical formalism [9], [10]. AR approaches can also be classified with respect to the employed sensor technology. We distinguish between the computer vision technologies, wearable sensors techniques and passive sensors techniques. Systems based on computer vision techniques [11], [12], usually have at their disposal highly informative data from which to extract the features necessary to the AR. However, there are many challenges with video based AR such as illumination variations, occlusion and background changes. Moreover, their usability in the context of smart homes for monitoring the activities of residents is questionable since several studies showed that people consider these solutions too intrusive [13]. The techniques based on wearable sensors represent a well explored research area [1], [14]. Initially, dedicated wearable motion sensors were used to recognize different physical activities [15]. In the recent years, there has been a shift towards mobile phones, since they are equipped with various sensors (GPS, accelerometer, gyroscope, etc.) and because they have become more powerful in terms of CPU, memory and battery. The technologies based on passive sensors can rely on a wide range of sensors: reed sensors, RFID tags, and PIR sensors are just some examples of acquisition devices field-tested [16], [17]. Novel AR systems based on power consumption readings also fall in this category [18], [19]. Whatever the system used, many real-world applications that focus on addressing needs of a human require information about the activities being performed by the users in real-time [20]–[22]. In this case the selected approach to perform AR must work in an online or streaming fashion and recognizing activities as and when new sensor events are recorded. There is the need for online AR techniques that can classify data as they are being collected. This is a challenging problem as data that completely describe an activity are not generally available in such situations and the algorithm has to rely on the partially observed data along with other contextual information to make a decision on the activity being performed. In [23], Wang et al. propose a real-time hierarchical model to recognize both simple gestures and complex activities using a wireless body sensor network. In this model, they first use a fast algorithm to detect gestures at the sensor node level, and