FFAME (Filtering Familiar Audio for Movement Exploration) is a novel sonification framework aiming to facilitate movement in individuals with chronic back pain. Our personalised, music-based approach contrasts and extends prior work with predetermined tonal sonification. FFAME progressively filters selected music based on angles of the trunk. Through a qualitative analysis of reported experience of 15 participants with chronic pain and 5 physiotherapists, we identify how sonification parameters and musical characteristics affect movement and meaning-making. Music-based movement sonification proved impactful across multiple dimensions: (1) encouraging movement, (2) escaping pain-related rumination, (3) externalizing pain experiences, and (4) scaffolding physical activities. Drawing on enactivism and related philosophies, the study highlights how the semantic indeterminacy of music, combined with real-time movement sonification, created a rich, open-ended environment that supported user agency and exploration. Sonification for pain management can be creative and expressive, enabling people with pain to extend challenging movements and build movement confidence.

Chronic pain is a prevalent condition where fear of movement and pain interfere with everyday functioning. Yet, there is no open body movement dataset for people with chronic pain in everyday settings. Our EmoPain@Home dataset addresses this with capture from 18 people with and without chronic pain in their homes, while they performed their routine activities. The data includes labels for pain, worry, and movement confidence continuously recorded for activity instances for the people with chronic pain. We explored baseline two-level pain detection based on this dataset and obtained 0.62 mean F1 score. However, extension of the dataset led to deterioration in performance confirming high variability in pain expressions for real world settings. We investigated baseline activity recognition for this setting as a first step in exploring the use of the activity label as contextual information for improving pain level classification performance. We obtained mean F1 score of 0.43 for 9 activity types, highlighting its feasibility. Further exploration, however, showed that data from healthy people cannot be easily leveraged for improving performance because worry and low confidence alter activity strategies for people with chronic pain. Our dataset and findings lay critical groundwork for automatic assessment of pain experience and behaviour in the wild.

Self-supervised learning has shown value for uncovering informative movement features for human activity recognition. However, there has been minimal exploration of this approach for affect recognition where availability of large labelled datasets is particularly limited. In this paper, we propose a P-STEMR (Parallel Space-Time Encoding Movement Representation) architecture with the aim of addressing this gap and specifically leveraging the higher availability of human activity recognition datasets for pain-level classification. We evaluated and analyzed the architecture using three different datasets across four sets of experiments. We found statistically significant increase in average F1 score to 0.84 for pain level classification with two classes based on the architecture compared with the use of hand-crafted features. This suggests that it is capable of learning movement representations and transferring these from activity recognition based on data captured in lab settings to classification of pain levels with messier real-world data. We further found that the efficacy of transfer between datasets can be undermined by dissimilarities in population groups due to impairments that affect movement behaviour and in motion primitives (e.g. rotation versus flexion). Future work should investigate how the effect of these differences could be minimized so that data from healthy people can be more valuable for transfer learning.

The use of multiple raters to label datasets is an established practice in affective computing. The principal goal is to reduce unwanted subjective bias in the labelling process. Unfortunately, this leads to the key problem of identifying a ground truth for training the affect recognition system. This problem becomes more relevant in a sparsely-crossed annotation where each rater only labels a portion of the full dataset to ensure a manageable workload per rater. In this paper, we introduce a Multi-Rater Consensus Learning (MRCL) method which learns a representative affect recognition model that accounts for each rater's agreement with the other raters. MRCL combines a multitask learning (MTL) regularizer and a consensus loss. Unlike standard MTL, this approach allows the model to learn to predict each rater's label while explicitly accounting for the consensus among raters. We evaluated our approach on two different datasets based on spontaneous affective body movement expressions for pain behaviour detection and laughter type recognition respectively. The two naturalistic datasets were chosen for the different forms of labelling (different in affect, observation stimuli, and raters) that they together offer for evaluating our approach. Empirical results demonstrate that MRCL is effective for modelling affect from datasets with sparsely-crossed multi-rater annotation.

Touch exploration of fabric is used to evaluate its properties, and it could further be leveraged to understand a consumer’s sensory experience and preference so as to support them in real time to make careful clothing purchase decisions. In this paper, we open up opportunities to explore the use of technology to provide such support with our FabricTouch dataset, i.e., a multimodal dataset of fabric assessment touch gestures. The dataset consists of bilateral forearm movement and muscle activity data captured while 15 people explored 114 different garments in total to evaluate them according to 5 properties (warmth, thickness, smoothness, softness, and flexibility). The dataset further includes subjective ratings of the garments with respect to each property and ratings of pleasure experienced in exploring the garment through touch. We further report baseline work on automatic detection. Our results suggest that it is possible to recognise the type of fabric property that a consumer is exploring based on their touch behaviour. We obtained mean F1 score of 0.61 for unseen garments, for 5 types of fabric property. The results also highlight the possibility of additionally recognizing the consumer’s subjective rating of the fabric when the property being rated is known, mean F1 score of 0.97 for unseen subjects, for 3 rating levels.

Given the importance of affective touch in human interactions, technology designers are increasingly attempting to bring this modality to the core of interactive technology. Advances in haptics and touch-sensing technology have been critical to fostering interest in this area. In this survey, we review how affective touch is investigated to enhance and support the human experience with or through technology. We explore this question across three different research areas to highlight their epistemology, main findings, and the challenges that persist. First, we review affective touch technology through the human–computer interaction literature to understand how it has been applied to the mediation of human–human interaction and its roles in other human interactions particularly with oneself, augmented objects/media, and affect-aware devices. We further highlight the datasets and methods that have been investigated for automatic detection and interpretation of affective touch in this area. In addition, we discuss the modalities of affective touch expressions in both humans and technology in these interactions. Second, we separately review how affective touch has been explored in human–robot and real-human–virtual-human interactions where the technical challenges encountered and the types of experience aimed at are different. We conclude with a discussion of the gaps and challenges that emerge from the review to steer research in directions that are critical for advancing affective touch technology and recognition systems. In our discussion, we also raise ethical issues that should be considered for responsible innovation in this growing area.

Chronic pain is a prevalent condition that affects everyday life of people around the world. Protective behaviors (strategies that are naturally but unhelpfully adopted by people with chronic pain to cope with fear of pain in executing harmless everyday movements) can lead to further disability over time if not recognized and addressed appropriately. In this paper, we build on previous work on unimodal, activity-independent, time-continuous protective behavior detection (PBD) by focusing on the fusion of muscle activity and body movement modalities for characterizing both protective behavior and its physical activity context. We explore different fusion strategies based on consideration of the manner in which protective behavior influences muscle activity and overt body movement as well as the relationship between the two modalities. We evaluate the various strategies on the multimodal EmoPain dataset containing data from people with and without chronic pain engaged in physical activities that reflect everyday challenges for those with chronic pain. Our results show that a central (model-level) fusion approach leads to better PBD performance than input- and decision-level fusions, or unimodal approaches. We also show that additional use of attention mechanism, typifying shifts in attention characteristic of protective behavior, further improves the sensitivity of the model, i.e. detection of the positive class (which is the minority class). We analyze these results and suggest that fusion in modelling a motor condition should consider how emotional responses (fear of movement and pain in this case) triggered by a condition affect each of the given modalities and hence their contributions to the modelling task.

While there is growing interest in developing tech-nology to support pain assessment, pain-related self-management, and healthcare personalisation, there are currently no datasets on nonverbal pain behaviour in the context of functional activities. To address this gap, we introduce the EmoPain(at)Home dataset which consists of motion capture data and self-reported pain, worry, and confidence intensities captured from people with chronic pain. The data were recorded during self-selected functional activities in the home, e.g. vacuuming. We include analysis of the dataset as well as baseline classification of pain levels with average F1 score of 0.61 for two classes. We additionally discuss inclusivity considerations for capture of datasets in naturalistic settings, based on lessons learnt within our study.

For technology (like serious games) that aims to deliver interactive learning, it is important to address relevant mental experiences such as reflective thinking during problem solving. To facilitate research in this direction, we present the weDraw-1 Movement Dataset of body movement sensor data and reflective thinking labels for 26 children solving mathematical problems in unconstrained settings where the body (full or parts) was required to explore these problems. Further, we provide qualitative analysis of behaviours that observers used in identifying reflective thinking moments in these sessions. The body movement cues from our compilation informed features that led to average F1 score of 0.73 for binary classification of problem-solving episodes by reflective thinking based on Long Short-Term Memory neural networks. We further obtained 0.79 average F1 score for end-to-end classification, i.e., based on raw sensor data. Finally, the algorithms resulted in 0.64 average F1 score for subsegments of these episodes as short as 4 seconds. Overall, our results show the possibility of detecting reflective thinking moments from body movement behaviours of a child exploring mathematical concepts bodily, such as within serious game play.

Comfort is a subjective experience that people attend to in everyday life including in cars where they are constrained in movement. Could intelligent cars sense their comfort levels for the purpose of maximizing it? To address this, first, we present a new dataset (available on request) of physical measures (skin temperature, blood volume pulse, electrodermal activity, and motion capture) and subjective thermal, sitting, and mental relaxation experience variables captured in semi-ecological settings in a car. Second, we provide an in-depth analysis of the relationship between passengers’ thermal experiences and physiological responses in the collected data. Our findings highlight complex duality in the relationship of thermal experience with heart rate variability and skin temperature variability. We discuss the practical implications that this may have for designing machine learning architectures for automatic detection of thermal discomfort.

In chronic pain rehabilitation, physiotherapists adapt physical activity to patients’ performance based on their expression of protective behavior, gradually exposing them to feared but harmless and essential everyday activities. As rehabilitation moves outside the clinic, technology should automatically detect such behavior to provide similar support. Previous works have shown the feasibility of automatic protective behavior detection (PBD) within a specific activity. In this article, we investigate the use of deep learning for PBD across activity types, using wearable motion capture and surface electromyography data collected from healthy participants and people with chronic pain. We approach the problem by continuously detecting protective behavior within an activity rather than estimating its overall presence. The best performance reaches mean F1 score of 0.82 with leave-one-subject-out cross validation. When protective behavior is modeled per activity type, performance achieves a mean F1 score of 0.77 for bend-down, 0.81 for one-leg-stand, 0.72 for sit-to-stand, 0.83 for stand-to-sit, and 0.67 for reach-forward. This performance reaches excellent level of agreement with the average experts’ rating performance suggesting potential for personalized chronic pain management at home. We analyze various parameters characterizing our approach to understand how the results could generalize to other PBD datasets and different levels of ground truth granularity.

The EmoPain 2020 Challenge is the first international competition aimed at creating a uniform platform for the comparison of multi-modal machine learning and multimedia processing methods of chronic pain assessment from human expressive behaviour, and also the identification of pain-related behaviours. The objective of the challenge is to promote research in the development of assistive technologies that help improve the quality of life for people with chronic pain via real-time monitoring and feedback to help manage their condition and remain physically active. The challenge also aims to encourage the use of the relatively underutilised, albeit vital bodily expression signals for automatic pain and pain-related emotion recognition. This paper presents a description of the challenge, competition guidelines, bench-marking dataset, and the baseline systems' architecture and performance on the Challenge's three sub-tasks: pain estimation from facial expressions, pain recognition from multimodal movement, and protective movement behaviour detection.

We propose a novel neural network architecture, named the Global Workspace Network (GWN), which addresses the challenge of dynamic and unspecified uncertainties in multimodal data fusion. Our GWN is a model of attention across modalities and evolving through time, and is inspired by the well-established Global Workspace Theory from the field of cognitive science. The GWN achieved average F1 score of 0.92 for discrimination between pain patients and healthy participants and average F1 score = 0.75 for further classification of three pain levels for a patient, both based on the multimodal EmoPain dataset captured from people with chronic pain and healthy people performing different types of exercise movements in unconstrained settings. In these tasks, the GWN significantly outperforms the typical fusion approach of merging by concatenation. We further provide extensive analysis of the behaviour of the GWN and its ability to address uncertainties (hidden noise) in multimodal data.

Clinicians tailor intervention in chronic pain rehabilitation to movement related self-efficacy (MRSE). This motivates us to investigate automatic MRSE estimation in this context towards the development of technology that is able to provide appropriate support in the absence of a clinician. We first explored clinical observer estimation, which showed that body movement behaviours, rather than facial expressions or engagement behaviours, were more pertinent to MRSE estimation during physical activity instances. Based on our findings, we built a system that estimates MRSE from bodily expressions and bodily muscle activity captured using wearable sensors. Our results (F1 scores of 0.95 and 0.78 in two physical exercise types) provide evidence of the feasibility of automatic MRSE estimation to support chronic pain physical rehabilitation. We further explored automatic estimation of MRSE with a reduced set of low-cost sensors to investigate the possibility of embedding such capabilities in ubiquitous wearable devices to support functional activity. Our evaluation for both exercise and functional activity resulted in F1 score of 0.79. This result suggests the possibility of (and calls for more studies on) MRSE estimation during everyday functioning in ubiquitous settings. We provide a discussion of the implication of our findings for relevant areas.

When developing interactive systems for children, such as serious games in the context of educational technology, it is important to take into account and address relevant cognitive and emotional child's experiences that may influence learning outcomes. Some works were done to analyze and automatically recognize these cognitive and affective states from nonverbal expressive behaviors. However, there is a lack of knowledge about visually impaired children and their body language to convey those states during learning tasks. In this paper, we present an analysis of nonverbal expressive behaviors of both blind and low-vision children, aiming at understanding what type of body communication can be an indicator of two cognitive states: engagement and confidence. In the study we consider the data collected along the EU-ICT H2020 weDRAW Project, while children were asked to solve mathematical tasks with their body. For such a dataset, we propose a list of 31 nonverbal behaviors, annotated both by rehabilitators used to work with visually impaired children and by naive observers. In the last part of the paper, we propose a preliminary study on automatic recognition of engagement and confidence states from 2D positional data. The classification results are up to 0.71 (F-score) on a three-class classification task.

In chronic pain physical rehabilitation, physiotherapists adapt exercise sessions according to the movement behavior of patients. As rehabilitation moves beyond clinical sessions, technology is needed to similarly assess movement behaviors and provide such personalized support. In this paper, as a first step, we investigate automatic detection of protective behavior (movement behavior due to pain-related fear or pain) based on wearable motion capture and electromyography sensor data. We investigate two recurrent networks (RNN) referred to as stacked-LSTM and dual-stream LSTM, which we compare with related deep learning (DL) architectures. We further explore data augmentation techniques and additionally analyze the impact of segmentation window lengths on detection performance. The leading performance of 0.815 mean F1 score achieved by stacked-LSTM provides important grounding for the development of wearable technology to support chronic pain physical rehabilitation during daily activities.

Although clinical best practice suggests that affect awareness could enable more effective technological support for physical rehabilitation through personalisation to psychological needs, designers need to consider what affective states matter, and how they should be tracked and addressed. In this article, we set the standard by analysing how the major affective factors in chronic pain (pain, fear/anxiety, and low/depressed mood) interfere with everyday physical functioning. Further, based on discussion of the modality that should be used to track these states to enable technology to address them, we investigated the possibility of using movement behaviour to automatically detect the states. Using two body movement datasets on people with chronic pain, we show that movement behaviour enables very good discrimination between two emotional distress levels (F1=0.86), and three pain levels (F1=0.9). Performance remained high (F1=0.78 for two pain levels) with a reduced set of movement sensors. Finally, in an overall discussion, we suggest how technology-provided encouragement and awareness can be personalised given the capability to automatically monitor the relevant states, towards addressing the barriers that they pose. In addition, we highlight movement behaviour features to be tracked to provide technology with information necessary for such personalisation.

People with chronic musculoskeletal pain would benefit from technology that provides run-time personalized feedback and help adjust their physical exercise plan. However, increased pain during physical exercise, or anxiety about anticipated pain increase, may lead to setback and intensified sensitivity to pain. Our study investigates the possibility of detecting pain levels from the quality of body movement during two functional physical exercises. By analyzing recordings of kinematics and muscle activity, our feature optimization algorithms and machine learning techniques can automatically discriminate between people with low level pain and high level pain and control participants while exercising. Best results were obtained from feature set optimization algorithms: 94% and 80% for the full trunk flexion and sit-to-stand movements respectively using Support Vector Machines. As depression can affect pain experience, we included participants' depression scores on a standard questionnaire and this improved discrimination between the control participants and the people with pain when Random Forests were used.

Touch is a primary nonverbal communication channel used to communicate emotions or other social messages. Despite its importance, this channel is still very little explored in the affective computing field, as much more focus has been placed on visual and aural channels. In this paper, we investigate the possibility to automatically discriminate between different social touch types. We propose five distinct feature sets for describing touch behaviours captured by a grid of pressure sensors. These features are then combined together by using the Random Forest and Boosting methods for categorizing the touch gesture type. The proposed methods were evaluated on both the HAART (7 gesture types over different surfaces) and the CoST (14 gesture types over the same surface) datasets made available by the Social Touch Gesture Challenge 2015. Well above chance level performances were achieved with a 67% accuracy for the HAART and 59% for the CoST testing datasets respectively.

Physical activity is essential in chronic pain rehabilitation. However, anxiety due to pain or a perceived exacerbation of pain causes people to guard against beneficial exercise. Interactive rehabiliation technology sensitive to such behaviour could provide feedback to overcome such psychological barriers. To this end, we developed a Support Vector Machine framework with the feature level fusion of body motion and muscle activity descriptors to discriminate three levels of pain (none, low and high). All subjects underwent a forward reaching exercise which is typically feared among people with chronic back pain. The levels of pain were categorized from control subjects (no pain) and thresholded self reported levels from people with chronic pain. Salient features were identified using a backward feature selection process. Using feature sets from each modality separately led to high pain classification F1 scores of 0.63 and 0.69 for movement and muscle activity respectively. However using a combined bimodal feature set this increased to F1 = 0.8.