Autism is a heterogenous condition, encompassing many different subtypes and presentations. Of those people with autism who lack communicative speech, some are more skilled at receptive language than their expressive difficulty might suggest. This disparity between what can be spoken and what can be understood correlates with motor and especially oral motor abilities, and thus may be a consequence of limits to oral motor skill. Point OutWords, tablet-based software targeted for this subgroup, builds on autistic perceptual and cognitive strengths to develop manual motor and oral motor skills prerequisite to communication by pointing or speaking. Although typical implementations of user-centered design rely on communicative speech, Point OutWords users were involved as co-creators both directly via their own nonverbal behavioral choices and indirectly via their communication therapists' reports; resulting features include vectorized, high-contrast graphics, exogenous cues to help capture and maintain attention, customizable reinforcement prompts, and accommodation of open-loop visuomotor control.

Wearable inertial measurement units incorporating accelerometers and gyroscopes are increasingly used for activity analysis and recognition. In this paper an activity classification algorithm is presented which includes a novel multi- step refinement with the aim of improving the classification accuracy of traditional approaches. To do so, after the classification takes place, information is extracted from the confusion matrix to focus the computational efforts on those activities with worse classification performance. It is argued that activities differ diversely from each other, therefore a specific set of features may be informative to classify a specific set of activities, but such informativeness should not necessarily be extended to a different activity set. This approach has shown promising results, achieving important classification accuracy improvements.

We present a segmentation algorithm capable of segmenting exercise repetitions in real time. This approach uses subsequence dynamic time warping and requires only a single exemplar repetition of an exercise to correctly segment repetitions from other subjects, including those with limited mobility. This approach is invariant to low range of motion, instability in movements, and sensor noise while remaining selective to different exercises. This algorithm enables responsive feedback for technology-assisted physical rehabilitation systems. We evaluated the algorithm against a publicly available dataset (CMU) and against a healthy population and stroke patient population performing rehabilitation exercises captured on a consumer-level depth sensor. We show that the algorithm can consistently achieve correct segmentation in real time.

Human Activities Recognition (HAR) based on low-level sensory data has become an active research topic and attracting attention in many application domains. Many approaches are employed to process and analyse the collected sensory data for modelling and representing Activity of Daily Working (ADW) and/or Activity of Daily Living (ADL). In this paper, a novel method based on Fuzzy Finite State Machine (FuFSM) is presented to model the daily activities. The proposed method is using FuFSM integrated with Fuzzy C-Means (FCMs) clustering algorithm to overcome the challenges of defining simultaneous activities. Therefore, different states of activities could be represented with a degree of fuzziness. Experimental results are presented to demonstrate the effectiveness of the proposed method. The model is tested and evaluated using a set of data that has been collected from an office environment.

Most recommendation systems for music rely on individual song ratings. Current song recommendation software that uses playlists has shown to either be inaccurate or suggest songs that are extremely like those in the playlist already. Furthermore, this recommendation software tends to rely very large numbers of records. AI models are used to overcome these limitations using substantially less data. A collaborative filtering approach using two different models (K-means and hierarchical clustering) is used to separate playlist data into clusters for comparison. After the data has been clustered, a Euclidean distance measure is used between the songs in the cluster and the average values of the songs in a single users playlist to make the final predictions. The use of normalisation and PCA enabled the K-means and hierarchical clustering models to form clusters efficiently. When tested on a small sample of users, the system recommended songs that were considered likeable by the users 60% of the time, while still finding songs that were generally diverse.

Encouraging rehabilitation by the use of technology in the home can be a cost-effective strategy, particularly if consumer-level equipment can be used. We present a clinical qualitative and quantitative analysis of the pose estimation algorithms of a typical consumer unit (Xbox One Kinect), to assess its suitability for technology supervised rehabilitation and guide development of future pose estimation algorithms for rehabilitation applications. We focused the analysis on upper-body stroke rehabilitation as a challenging use case. We found that the algorithms require improved joint tracking, especially for the shoulder, elbow and wrist joints, and exploiting temporal information for tracking when there is full or partial occlusion in the depth data.

In this paper, a novel approach to the container loading problem using a spatial entropy measure to bias a Monte Carlo Tree Search is proposed. The proposed algorithm generates layouts that achieve the goals of both fitting a constrained space and also having consistency or neatness that enables forklift truck drivers to apply them easily to real shipping containers loaded from one end. Three algorithms are analysed. The first is a basic Monte Carlo Tree Search, driven only by the principle of minimising the length of container that is occupied. The second is an algorithm that uses the proposed entropy measure to drive an otherwise random process. The third algorithm combines these two principles and produces superior results to either. These algorithms are then compared to a classical deterministic algorithm. It is shown that where the classical algorithm fails, the entropy-driven algorithms are still capable of providing good results in a short computational time.

One of the imminent challenges for assistive robots in learning human activities while observing a human perform a task is how to define movement representations (states). This has been recently explored for improved solutions. This paper proposes a method of extracting key frames (or poses) of human activities from skeleton joint coordinates information obtained using an RGB-D Camera (Depth Sensor). The motion energy (kinetic energy) of each pose in an activity sequence is computed and a novel approach is proposed for extracting key pose locations that define an activity using moving average crossovers of computed pose kinetic energy. This is important as not all frames of an activity sequence are key in defining the activity. In order to evaluate the reliability of extracted key poses, Long Short-Term Memory (LSTM) Recurrent Neural Network (RNN) which is capable to learn a sequence of transition from states in an activity is applied in classifying activities from identified key poses. This is important for assistive robots to identify key human poses and states transition in order to correctly carry out human activities. Some preliminary experimental results are presented to illustrate the proposed methodology.

In this paper, a clustering-based fuzzy finite state machine approach for human activity modelling and recognition is proposed. It Incorporates the Fuzzy C-means (FCMs) clustering algorithm with a Fuzzy Finite State Machine (FuFSM) in order to generate the state transitions more effectively. This unsupervised approach will overcome the deficiency in identifying the knowledge-base required for FuFSM. To validate the proposed approach, experimental results are presented. The activities of two office workers are modelled/recognised using the proposed method. The approach taken for this research is based on ambient Intelligent sensory data rather than data coming from wearable sensors.

Falls are one of the greatest risks for older adults living alone at home. This paper presents a novel visual-based fall detection approach to support independent living for older adults through analysing the motion and shape of the human body. The proposed approach employs a new set of features to detect a fall. Motion information of a segmented silhouette when extracted can provide a useful cue for classifying different behaviours, while variation in shape and the projection histogram can be used to describe human body postures and subsequent fall events. The proposed approach presented here extracts motion information using best-fit approximated ellipse and bounding box around the human body, produces projection histograms and determines the head position over time, to generate 10 features to identify falls. These features are fed into a multilayer perceptron neural network for fall classification. Experimental results show the reliability of the proposed approach with a high fall detection rate of 99.60% and a low false alarm rate of 2.62% when tested with the UR Fall Detection dataset. Comparisons with state of the art fall detection techniques show the robustness of the proposed approach.

Despite the increasing attention given to inertial sensors for Human Activity Recognition (HAR), efforts are principally focused on fitness applications where quasi-periodic activities like walking or running are studied. In contrast, activities like eating or drinking cannot be considered periodic or quasi-periodic. Instead, they are composed of sporadic occurring gestures in continuous data streams. This paper presents an approach to gesture recognition for an Ambient Assisted Living (AAL) environment. Specifically, food and drink intake gestures are studied. To do so, firstly, waist-worn tri-axial accelerometer data is used to develop a low computational model to recognize whether a person is at moving, sitting or standing estate. With this information, data from a wrist-worn tri-axial Micro-Electro-Mechanical (MEM) system was used to recognize a set of similar eating and drinking gestures. The promising preliminary results show that states can be recognized with 100% classification accuracy with the use of a low computational model on a reduced 4-dimensional feature vector. Additionally, the recognition rate achieved for eating and drinking gestures was above 99%. Altogether suggests that it is possible to develop a continuous monitoring system based on a bi-nodal inertial unit. This work is part of a bigger project that aims at developing a self-neglect detection continuous monitoring system for older adults living independently.

Physical activities have tremendous benefit to older adults. A report from the World Health Organization has mentioned that lack of physical activity contributed to around 3.2 million premature deaths annually worldwide. Research also shows that regular exercise helps the older adults by improving their physical fitness, immune system, sleep and stress levels, not to mention the countless health problems it reduces such as diabetes, cardiovascular disease, dementia, obesity, joint pains, etc. The research reported in this paper is introducing a Socially Assistive Robot (SAR) that will engage, coach, assess and motivate the older adults in physical exercises that are recommended by the National Health Services (NHS) in the UK. With the rise in the population of older adults, which is expected to triple by 2050, this SAR will aim to improve the quality of life for a significant proportion of the population. To assess the proposed robot exercise trainer, user's observational evaluation with 17 participants is conducted. Participants are generally happy with the proposed platform as a mean of encouraging them to do regular exercise correctly.

Activities of daily living (ADL) or activities of daily working (ADW) may be affected by changes in a person's health or well-being. Measuring progressive changes in one activity or multiple activities is representative of behavioural variations. By inspecting the trends in multiple activities, it is possible to identify and predict human behavioural changes. We refer to the trends in people's behaviour as behavioural evolution. In this paper, we propose a novel indicator to measure the progressive changes representing a participant's behavioural evolution. The proposed indicator presents activities as a holistic measure, which first combine multi-activities and then measure the progressive changes in the combined activities for each single day. Real data sets were collected from a wireless sensor network and used to examine our proposed technique. As part of this process, we were able to quantify progressive changes for individual and aggregated activities. Our experimental results demonstrated that: (1) the proposed approach can identify and distinguish normal and abnormal behaviours; (2) large data sets gathered from sensors in an intelligent environment represented in various time series can be visualised in a simple and more understandable format; (3) identifying trends in ADLs or ADWs is a relevant means of sharing information with carers or supervisors. (C) 2018 Elsevier Ltd. All rights reserved.

Abu Khari A'ain Ghulam Abbas Mohd Zaidi Abd Rozan Normaziah Abdul Aziz Izhal Abdul Halin Ruzairi Abdul Rahim Rosni Abdulla Dayang Norhayati Abg Jawawi Hassan Abolhassani Ajith Abraham Kamalrulnizam Abu Bakar Rohani Abu Bakar Syed Abd Rahman Abu Bakar Khalid Al-Begain David Al-Dabass Dhiya Al-Jumeily Mikulas Alexik Belal Alhaija Tony Allen Ferda Alpaslan Shamsudin Amin Konar Amit Obinna Anya Irfan Awan Irfan Awan Eduard Babulak Kambiz Badie Gurvinder Baicher Gurvinder-Singh Baicher Preeti Bajaj Narendra Bawane Fabian Boettinger Felix Breitenecker Adam Brentnall Agostino Bruzzone Piers Campbell Richard Cant Andre Carvalho Sanjay Chaudhary Russell Cheng Monica Chis Roy Crosbie Jiri Dvorsky Mazlina Esa G Ganesan Jafar Habibi Fazilah Haron Habibollah Haron Manaf Hashim Vlatka Hlupic Zuwairie Ibrahim Mohd. Yazid Idris Teruaki Ito Janos-Sebestyen Janosy Emilio Macias Esko Juuso Helen Karatza S. D. Katebi Marzuki Khalid Hisham Khamis Shubha Kher Petia Koprinkova Jiri Kunovsky Caroline Langensiepen Malcolm Yoke Hean Low Mahdi Mahfouf Khalid Marzuki Yuri Merkuryev Galina Merkuryeva Radziah Mohamad Samsul Bahari Mohd-Noor Atulya Nagar Gaby Neumann Lars Nolle Alessandra Orsoni Taha Osman Mohd Fauzi Othman Charles Patchett Gillian Pearce Mirjana Pejic-Bach Evtim Peytchev Heather Powell Steve Presland Marco Remondino Norlaili Safri Zaliman Sauli Siti Mariyam Shamsuddin Kumbakonam Govindarajan Subramanian Hissam Tawfik Palaniappagounder Thangavel Patrick Wang Wolfgang Wiechert Fadi Yaacoub Jasmy Yunus Suiping Zhou Richard Zobel Borut Zupančič

Human activity recognition (HAR) has mainly been directed to the recognition of static or quasi-periodic activities like sitting, walking or running, typically for fitness applications. However, activities like eating or drinking are neither static nor quasi-periodic. Instead, they are composed of sparsely occurring motions or gestures in continuous data streams. This paper presents a novel adaptive segmentation technique based on crosses of moving averages to identify potential eating or drinking gestures from accelerometer data. The novel crossings-based segmentation approach proposed is able to identify all eating and drinking gestures from continuous accelerometer data including different activities. A posteriori, potential gestures are classified as food or drink intake gestures using a combination of Dynamic Time Warping (DTW) as signal similarity measure and a k-Nearest Neighbours (KNN) classifier. An outstanding classification rate of 100% has been achieved.

Different computational methodologies for anomaly detection has been studied in the past. Novelty detection involves classifying if test data differs from the training data. This is applicable to a scenario when there are sufficiently many normal training samples and little or no abnormal data. In this research, a novelty detection algorithm known as One-Class Support Vector Machine (SVM) is applied for detection of anomaly in Activities of Daily Living (ADL), specifically sleeping patterns, which could be a sign of Mild Cognitive Impairment (MCI) in older adults or other health-related issues. Tests conducted on both synthetic and real data shows promising results.

Analysis of human behaviour changes is a subject of interest for many researchers. This could be obtained considering either short-term or long-term changes. The aim of this study is to find long-term changes (behaviour evolution) in Activities of Daily Living (ADL) or Activities of Daily Working (ADW) of users in an Ambient Intelligence (AmI) environment. Analysis is based on introduction of a novel Human Behaviour Momentum Indicator (HBMI). Extensive experiments are conducted to investigate the effectiveness of the studied techniques on real-world datasets collected from home and office environments. To show the effectiveness of the proposed approach, results are compared with Relative Strength Index (RSI). The results show that trends in ADL or ADW can be detected and the direction of the activity's trend are predicted. In addition, the results show that our proposed technique gives a better response to changes in data more than the other technique.

Family care is the most accepted and preferred care setting for both long-term care patients and their relatives. However, many of these caregivers are elderly people themselves, and often reach the point where they also need support. Care poses a substantial burden, so often it is not the health of the patient but the overload of stress for the caregiver that results in the need for much more expensive professional care and even residential care. An ambient assisted living technology platform is developed to support both older adults and their carers to overcome the challenges of the care. The platform offers informal carers support by means of monitoring Activities of Daily Care as well as their psychological state, and will provide orientation to help them improve the care given. Monitored information will be registered by means of home-installed and personal sensor technologies based on Internet of Things (IoT), which will be as unobtrusive as possible for the house inhabitants.

An ideal binary mask is a means by which multiple sound sources within a single audio file can be separated. Previous work has shown a deep neural network can be trained to approximate the ideal mask, but at a substantial computational cost. We present a method to assess the impact of reducing the mask by averaging time and frequency bins, so that the computational cost can be significantly reduced. Our work uses the original separate musical channels mask as a ground truth and compares this against an ideal binary mask and an ideal ”soft” or proportional mask. The ideal soft mask is then compared against masks produced by a range of averaging levels. We find that averaging could produce a reduction by a factor of 16 in the number of weights in the neural network (and thus a significant improvement in computation time), while still achieving plausible results in terms of source separation.