Identificazione di attività motorie a partire da misure con sensori indossabili: una revisione critica della letteratura

La revisione qui riportata valuta tutte le modalità di identificazione di task motori e posturali attraverso l'uso di sensori indossabili, principalmente accelerometri. Essa ha lo scopo di illustrare i sensori e gli algoritmi utilizzati in 23 articoli scelti in base alla loro qualità secondo una metodologia personalizzata di ricerca per fare il punto degli studi in questo campo, fino a questo momento. I dati estratti vengono utilizzati per individuare gli aspetti chiave riportati negli articoli, specialmente riguardanti l'algoritmo, focus della nostra revisione. Secondo questo criterio vengono selezionati 13 articoli, i quali si soffermano maggiormente sui modelli di approccio utilizzati, al fine di ottenere la più elevata accuratezza nell'identificazione. Questa in generale varia tra l'80-90% per i task motori più conosciuti(camminata, corsa e altri) mentre rimane limitata, intorno al 60-70% quando vengono analizzati i movimenti specifici degli arti superiori o inferiori.

Localization and activity detection Based on the fusion of environment And inertial sensors = Localización y detección de actividad Basadas en integración de sensores y dispositivos inerciales

Los sensores inerciales (acelerómetros y giróscopos) se han ido introduciendo poco a poco en dispositivos que usamos en nuestra vida diaria gracias a su minituarización. Hoy en día todos los smartphones contienen como mínimo un acelerómetro y un magnetómetro, siendo complementados en losmás modernos por giróscopos y barómetros. Esto, unido a la proliferación de los smartphones ha hecho viable el diseño de sistemas basados en las medidas de sensores que el usuario lleva colocados en alguna parte del cuerpo (que en un futuro estarán contenidos en tejidos inteligentes) o los integrados en su móvil. El papel de estos sensores se ha convertido en fundamental para el desarrollo de aplicaciones contextuales y de inteligencia ambiental. Algunos ejemplos son el control de los ejercicios de rehabilitación o la oferta de información referente al sitio turístico que se está visitando. El trabajo de esta tesis contribuye a explorar las posibilidades que ofrecen los sensores inerciales para el apoyo a la detección de actividad y la mejora de la precisión de servicios de localización para peatones. En lo referente al reconocimiento de la actividad que desarrolla un usuario, se ha explorado el uso de los sensores integrados en los dispositivos móviles de última generación (luz y proximidad, acelerómetro, giróscopo y magnetómetro). Las actividades objetivo son conocidas como ‘atómicas’ (andar a distintas velocidades, estar de pie, correr, estar sentado), esto es, actividades que constituyen unidades de actividades más complejas como pueden ser lavar los platos o ir al trabajo. De este modo, se usan algoritmos de clasificación sencillos que puedan ser integrados en un móvil como el Naïve Bayes, Tablas y Árboles de Decisión. Además, se pretende igualmente detectar la posición en la que el usuario lleva el móvil, no sólo con el objetivo de utilizar esa información para elegir un clasificador entrenado sólo con datos recogidos en la posición correspondiente (estrategia que mejora los resultados de estimación de la actividad), sino también para la generación de un evento que puede producir la ejecución de una acción. Finalmente, el trabajo incluye un análisis de las prestaciones de la clasificación variando el tipo de parámetros y el número de sensores usados y teniendo en cuenta no sólo la precisión de la clasificación sino también la carga computacional. Por otra parte, se ha propuesto un algoritmo basado en la cuenta de pasos utilizando informaiii ción proveniente de un acelerómetro colocado en el pie del usuario. El objetivo final es detectar la actividad que el usuario está haciendo junto con la estimación aproximada de la distancia recorrida. El algoritmo de cuenta pasos se basa en la detección de máximos y mínimos usando ventanas temporales y umbrales sin requerir información específica del usuario. El ámbito de seguimiento de peatones en interiores es interesante por la falta de un estándar de localización en este tipo de entornos. Se ha diseñado un filtro extendido de Kalman centralizado y ligeramente acoplado para fusionar la información medida por un acelerómetro colocado en el pie del usuario con medidas de posición. Se han aplicado también diferentes técnicas de corrección de errores como las de velocidad cero que se basan en la detección de los instantes en los que el pie está apoyado en el suelo. Los resultados han sido obtenidos en entornos interiores usando las posiciones estimadas por un sistema de triangulación basado en la medida de la potencia recibida (RSS) y GPS en exteriores. Finalmente, se han implementado algunas aplicaciones que prueban la utilidad del trabajo desarrollado. En primer lugar se ha considerado una aplicación de monitorización de actividad que proporciona al usuario información sobre el nivel de actividad que realiza durante un período de tiempo. El objetivo final es favorecer el cambio de comportamientos sedentarios, consiguiendo hábitos saludables. Se han desarrollado dos versiones de esta aplicación. En el primer caso se ha integrado el algoritmo de cuenta pasos en una plataforma OSGi móvil adquiriendo los datos de un acelerómetro Bluetooth colocado en el pie. En el segundo caso se ha creado la misma aplicación utilizando las implementaciones de los clasificadores en un dispositivo Android. Por otro lado, se ha planteado el diseño de una aplicación para la creación automática de un diario de viaje a partir de la detección de eventos importantes. Esta aplicación toma como entrada la información procedente de la estimación de actividad y de localización además de información almacenada en bases de datos abiertas (fotos, información sobre sitios) e información sobre sensores reales y virtuales (agenda, cámara, etc.) del móvil. Abstract Inertial sensors (accelerometers and gyroscopes) have been gradually embedded in the devices that people use in their daily lives thanks to their miniaturization. Nowadays all smartphones have at least one embedded magnetometer and accelerometer, containing the most upto- date ones gyroscopes and barometers. This issue, together with the fact that the penetration of smartphones is growing steadily, has made possible the design of systems that rely on the information gathered by wearable sensors (in the future contained in smart textiles) or inertial sensors embedded in a smartphone. The role of these sensors has become key to the development of context-aware and ambient intelligent applications. Some examples are the performance of rehabilitation exercises, the provision of information related to the place that the user is visiting or the interaction with objects by gesture recognition. The work of this thesis contributes to explore to which extent this kind of sensors can be useful to support activity recognition and pedestrian tracking, which have been proven to be essential for these applications. Regarding the recognition of the activity that a user performs, the use of sensors embedded in a smartphone (proximity and light sensors, gyroscopes, magnetometers and accelerometers) has been explored. The activities that are detected belong to the group of the ones known as ‘atomic’ activities (e.g. walking at different paces, running, standing), that is, activities or movements that are part of more complex activities such as doing the dishes or commuting. Simple, wellknown classifiers that can run embedded in a smartphone have been tested, such as Naïve Bayes, Decision Tables and Trees. In addition to this, another aim is to estimate the on-body position in which the user is carrying the mobile phone. The objective is not only to choose a classifier that has been trained with the corresponding data in order to enhance the classification but also to start actions. Finally, the performance of the different classifiers is analysed, taking into consideration different features and number of sensors. The computational and memory load of the classifiers is also measured. On the other hand, an algorithm based on step counting has been proposed. The acceleration information is provided by an accelerometer placed on the foot. The aim is to detect the activity that the user is performing together with the estimation of the distance covered. The step counting strategy is based on detecting minima and its corresponding maxima. Although the counting strategy is not innovative (it includes time windows and amplitude thresholds to prevent under or overestimation) no user-specific information is required. The field of pedestrian tracking is crucial due to the lack of a localization standard for this kind of environments. A loosely-coupled centralized Extended Kalman Filter has been proposed to perform the fusion of inertial and position measurements. Zero velocity updates have been applied whenever the foot is detected to be placed on the ground. The results have been obtained in indoor environments using a triangulation algorithm based on RSS measurements and GPS outdoors. Finally, some applications have been designed to test the usefulness of the work. The first one is called the ‘Activity Monitor’ whose aim is to prevent sedentary behaviours and to modify habits to achieve desired objectives of activity level. Two different versions of the application have been implemented. The first one uses the activity estimation based on the step counting algorithm, which has been integrated in an OSGi mobile framework acquiring the data from a Bluetooth accelerometer placed on the foot of the individual. The second one uses activity classifiers embedded in an Android smartphone. On the other hand, the design of a ‘Travel Logbook’ has been planned. The input of this application is the information provided by the activity and localization modules, external databases (e.g. pictures, points of interest, weather) and mobile embedded and virtual sensors (agenda, camera, etc.). The aim is to detect important events in the journey and gather the information necessary to store it as a journal page.

Wearable sensors in Huntington disease:a pilot study

Background: The Unified Huntington’s Disease Rating Scale (UHDRS) is the principal means of assessing motor impairment in Huntington disease but is subjective and generally limited to in-clinic assessments. Objective: To evaluate the feasibility and ability of wearable sensors to measure motor impairment in individuals with Huntington disease in the clinic and at home. Methods: Participants with Huntington disease and controls were asked to wear five accelerometer-based sensors attached to the chest and each limb for standardized, in-clinic assessments and for one day at home. A secondchest sensor was worn for six additional days at home. Gait measures were compared between controls, participants with Huntington disease, and participants with Huntington disease grouped by UHDRS total motor score using Cohen’s d values. Results: Fifteen individuals with Huntington disease and five controls completed the study. Sensor data were successfully captured from 18 of the 20 participants at home. In the clinic, the standard deviation of step time (timebetween consecutive steps) was increased in Huntington disease (p<0.0001; Cohen’s d=2.61) compared to controls. At home with additional observations, significant differences were observed in seven additional gait measures. The gait of individuals with higher total motor scores (50 or more) differed significantly from those with lower total motor scores (below 50) on multiple measures at home. Conclusions: In this pilot study, the use of wearable sensors in clinic and at home was feasible and demonstrated gait differences between controls, participants with Huntington disease, and participants with Huntington diseasegrouped by motor impairment.

Cascading appearance-based features for visual voice activity detection

The detection of voice activity is a challenging problem, especially when the level of acoustic noise is high. Most current approaches only utilise the audio signal, making them susceptible to acoustic noise. An obvious approach to overcome this is to use the visual modality. The current state-of-the-art visual feature extraction technique is one that uses a cascade of visual features (i.e. 2D-DCT, feature mean normalisation, interstep LDA). In this paper, we investigate the effectiveness of this technique for the task of visual voice activity detection (VAD), and analyse each stage of the cascade and quantify the relative improvement in performance gained by each successive stage. The experiments were conducted on the CUAVE database and our results highlight that the dynamics of the visual modality can be used to good effect to improve visual voice activity detection performance.

The QUT-NOISE-TIMIT corpus for the evaluation of voice activity detection algorithms

The QUT-NOISE-TIMIT corpus consists of 600 hours of noisy speech sequences designed to enable a thorough evaluation of voice activity detection (VAD) algorithms across a wide variety of common background noise scenarios. In order to construct the final mixed-speech database, a collection of over 10 hours of background noise was conducted across 10 unique locations covering 5 common noise scenarios, to create the QUT-NOISE corpus. This background noise corpus was then mixed with speech events chosen from the TIMIT clean speech corpus over a wide variety of noise lengths, signal-to-noise ratios (SNRs) and active speech proportions to form the mixed-speech QUT-NOISE-TIMIT corpus. The evaluation of five baseline VAD systems on the QUT-NOISE-TIMIT corpus is conducted to validate the data and show that the variety of noise available will allow for better evaluation of VAD systems than existing approaches in the literature.

The delta-phase spectrum with application to voice activity detection and speaker recognition

For several reasons, the Fourier phase domain is less favored than the magnitude domain in signal processing and modeling of speech. To correctly analyze the phase, several factors must be considered and compensated, including the effect of the step size, windowing function and other processing parameters. Building on a review of these factors, this paper investigates a spectral representation based on the Instantaneous Frequency Deviation, but in which the step size between processing frames is used in calculating phase changes, rather than the traditional single sample interval. Reflecting these longer intervals, the term delta-phase spectrum is used to distinguish this from instantaneous derivatives. Experiments show that mel-frequency cepstral coefficients features derived from the delta-phase spectrum (termed Mel-Frequency delta-phase features) can produce broadly similar performance to equivalent magnitude domain features for both voice activity detection and speaker recognition tasks. Further, it is shown that the fusion of the magnitude and phase representations yields performance benefits over either in isolation.

Noise robust voice activity detection using normal probability testing and time-domain histogram analysis

This paper presents a method of voice activity detection (VAD) suitable for high noise scenarios, based on the fusion of two complementary systems. The first system uses a proposed non-Gaussianity score (NGS) feature based on normal probability testing. The second system employs a histogram distance score (HDS) feature that detects changes in the signal through conducting a template-based similarity measure between adjacent frames. The decision outputs by the two systems are then merged using an open-by-reconstruction fusion stage. Accuracy of the proposed method was compared to several baseline VAD methods on a database created using real recordings of a variety of high-noise environments.

Noise robust voice activity detection using features extracted from the time-domain autocorrelation function

This paper presents a method of voice activity detection (VAD) for high noise scenarios, using a noise robust voiced speech detection feature. The developed method is based on the fusion of two systems. The first system utilises the maximum peak of the normalised time-domain autocorrelation function (MaxPeak). The second zone system uses a novel combination of cross-correlation and zero-crossing rate of the normalised autocorrelation to approximate a measure of signal pitch and periodicity (CrossCorr) that is hypothesised to be noise robust. The score outputs by the two systems are then merged using weighted sum fusion to create the proposed autocorrelation zero-crossing rate (AZR) VAD. Accuracy of AZR was compared to state of the art and standardised VAD methods and was shown to outperform the best performing system with an average relative improvement of 24.8% in half-total error rate (HTER) on the QUT-NOISE-TIMIT database created using real recordings from high-noise environments.

Visual voice activity detection using frontal versus profile views

Visual activity detection of lip movements can be used to overcome the poor performance of voice activity detection based solely in the audio domain, particularly in noisy acoustic conditions. However, most of the research conducted in visual voice activity detection (VVAD) has neglected addressing variabilities in the visual domain such as viewpoint variation. In this paper we investigate the effectiveness of the visual information from the speaker’s frontal and profile views (i.e left and right side views) for the task of VVAD. As far as we are aware, our work constitutes the first real attempt to study this problem. We describe our visual front end approach and the Gaussian mixture model (GMM) based VVAD framework, and report the experimental results using the freely available CUAVE database. The experimental results show that VVAD is indeed possible from profile views and we give a quantitative comparison of VVAD based on frontal and profile views The results presented are useful in the development of multi-modal Human Machine Interaction (HMI) using a single camera, where the speaker’s face may not always be frontal.

Process monitoring using sensors in YAWL

This article describes the architecture of a monitoring component for the YAWL system. The architecture proposed is based on sensors and it is realized as a YAWL service to have perfect integration with the YAWL systems. The architecture proposed is generic and applicable in different contexts of business process monitoring. Finally, it was tested and evaluated in the context of risk monitoring for business processes.

Telomerase activity : detection and clinical implications in human cancer

Telomerase is an extremely important enzyme required for the immortalisation of tumour cells. Because the gene is activated in the vast majority of tumour tissues and remains unused in most somatic cells, it represents a marker with huge diagnostic, prognostic and treatment implications in cancer. This article summarises the basic structure and functions of telomerase and considers its clinical implications in colorectal and other cancers.

Complete-linkage clustering for voice activity detection in audio and visual speech

We propose a novel technique for conducting robust voice activity detection (VAD) in high-noise recordings. We use Gaussian mixture modeling (GMM) to train two generic models; speech and non-speech. We then score smaller segments of a given (unseen) recording against each of these GMMs to obtain two respective likelihood scores for each segment. These scores are used to compute a dissimilarity measure between pairs of segments and to carry out complete-linkage clustering of the segments into speech and non-speech clusters. We compare the accuracy of our method against state-of-the-art and standardised VAD techniques to demonstrate an absolute improvement of 15% in half-total error rate (HTER) over the best performing baseline system and across the QUT-NOISE-TIMIT database. We then apply our approach to the Audio-Visual Database of American English (AVDBAE) to demonstrate the performance of our algorithm in using visual, audio-visual or a proposed fusion of these features.

Robust Whisper Activity Detection Using Long-Term Log Energy Variation of Sub-Band Signal

The goal in the whisper activity detection (WAD) is to find the whispered speech segments in a given noisy recording of whispered speech. Since whispering lacks the periodic glottal excitation, it resembles an unvoiced speech. This noise-like nature of the whispered speech makes WAD a more challenging task compared to a typical voice activity detection (VAD) problem. In this paper, we propose a feature based on the long term variation of the logarithm of the short-time sub-band signal energy for WAD. We also propose an automatic sub-band selection algorithm to maximally discriminate noisy whisper from noise. Experiments with eight noise types in four different signal-to-noise ratio (SNR) conditions show that, for most of the noises, the performance of the proposed WAD scheme is significantly better than that of the existing VAD schemes and whisper detection schemes when used for WAD.