Using clustering techniques for intelligent camera-based user interfaces

The area of Human-Machine Interface is growing fast due to its high importance in all technological systems. The basic idea behind designing human-machine interfaces is to enrich the communication with the technology in a natural and easy way. Gesture interfaces are a good example of transparent interfaces. Such interfaces must identify properly the action the user wants to perform, so the proper gesture recognition is of the highest importance. However, most of the systems based on gesture recognition use complex methods requiring high-resource devices. In this work, we propose to model gestures capturing their temporal properties, which significantly reduce storage requirements, and use clustering techniques, namely self-organizing maps and unsupervised genetic algorithm, for their classification. We further propose to train a certain number of algorithms with different parameters and combine their decision using majority voting in order to decrease the false positive rate. The main advantage of the approach is its simplicity, which enables the implementation using devices with limited resources, and therefore low cost. The testing results demonstrate its high potential.

Evaluación de las librerías disponibles para la definición de gestos y uso de los mismos para ser usados en U.I. con Kinect

La rápida evolución de tecnologías que mejoran la interacción con los ordenadores ha cambiado la forma de pensar de la sociedad actual. Una de las últimas alternativas se centra en el reconocimiento facial y de gestos, el cual ha dado muy buenos resultados, acercando ordenadores y humanos. El objetivo de este proyecto era desarrollar un prototipo que permita el reconocimiento de gestos mediante el dispositivo Kinect, el cual usa una de las tecnologías mencionadas anteriormente. Para la realización de este prototipo, fue necesario el estudio del funcionamiento de Kinect. Una vez realizado, el siguiente paso consistió en la búsqueda de librerías que facilitaran el reconocimiento de gestos, para compararlas y finalmente escoger una de ellas. Al ser una tecnología reciente, no existen demasiadas aplicaciones que utilicen Kinect y por lo tanto, la búsqueda de la librería era una parte fundamental del proyecto. Tras elegir la librería, se analizaron las características que debía tener el prototipo para realizar el diseño del mismo. Este diseño se realizó de forma iterativa y con numerosos cambios para otorgar al prototipo la mayor abstracción posible y una mayor orientación a objetos, para de esta manera, favorecer los posibles cambios que sean necesarios al ser una tecnología nueva y muy cambiante. Finalmente, el paso final fue la implementación del prototipo bajo el diseño propuesto, consiguiendo un prototipo funcional capaz de reconocer, almacenar, mostrar y definir diferentes gestos. Los resultados ofrecidos por Kinect son muy satisfactorios, con una gran precisión y eficiencia, demostrando el potencial de las nuevas tecnologías de interacción sin contacto, cuyo único defecto se encuentra en ser una tecnología poco madura y en evolución. ---------------------------------------------------------------------------------------------------------- The fast development of technologies that enhance interaction with computers has changed the mindset of the society. One of the latest alternatives focuses on face and gesture recognition, which has been very successful, bringing closer computers and humans. The purpose of this project was to develop a gesture recognition prototype with the Kinect device, which uses one of the technologies mentioned above. For the realization of this prototype, it was necessary to study how Kinect works. Once done, the next step was to search for libraries to facilitate the gesture recognition, to compare them and finally choose one of them. Because it is a recent technology, there are not many applications that use Kinect and therefore, the library search was an essential part of the project. After selecting the library, the characteristics that the prototype should have were analyzed for the design. This design was performed iteratively with numerous changes to give the prototype abstraction and object orientation, because as a new technology, it changes fast. Therefore, with this design it would be easier to deal with possible changes. Finally, the final step was the implementation of the prototype under the proposed design, getting a functional prototype able to recognize, store, display and define different gestures. The results offered by Kinect are very satisfactory, with high accuracy and efficiency, demonstrating the potential of the new interaction technologies without contact, whose only fault is to be a little mature technology and evolving.

Reconocimiento de gestos basado en acelerómetros

En los últimos años, ha crecido de forma significativa el interés por la utilización de dispositivos capaces de reconocer gestos humanos. En este trabajo, se pretenden reconocer gestos manuales colocando sensores en la mano de una persona. El reconocimiento de gestos manuales puede ser implementado para diversos usos y bajo diversas plataformas: juegos (Wii), control de brazos robóticos, etc. Como primer paso, se realizará un estudio de las actuales técnicas de reconocimiento de gestos que utilizan acelerómetros como sensor de medida. En un segundo paso, se estudiará como los acelerómetros pueden utilizarse para intentar reconocer los gestos que puedan realizar una persona (mover el brazo hacia un lado, girar la mano, dibujar un cuadrado, etc.) y los problemas que de su utilización puedan derivarse. Se ha utilizado una IMU (Inertial Measurement Unit) como sensor de medida. Está compuesta por tres acelerómetros y tres giróscopos (MTi-300 de Xsens). Con las medidas que proporcionan estos sensores se realiza el cálculo de la posición y orientación de la mano, representando esta última en función de los ángulos de Euler. Un aspecto importante a destacar será el efecto de la gravedad en las medidas de las aceleraciones. A través de diversos cálculos y mediante la ayuda de los giróscopos se podrá corregir dicho efecto. Por último, se desarrollará un sistema que identifique la posición y orientación de la mano como gestos reconocidos utilizando lógica difusa. Tanto para la adquisición de las muestras, como para los cálculos de posicionamiento, se ha desarrollado un código con el programa Matlab. También, con este mismo software, se ha implementado un sistema de lógica difusa con la que se realizará el reconocimiento de los gestos, utilizando la herramienta FIS Editor. Las pruebas realizadas han consistido en la ejecución de nueve gestos por diferentes personas teniendo una tasa de reconocimiento comprendida entre el 90 % y 100 % dependiendo del gesto a identificar. ABSTRACT In recent years, it has grown significantly interest in the use of devices capable of recognizing human gestures. In this work, we aim to recognize hand gestures placing sensors on the hand of a person. The recognition of hand gestures can be implemented for different applications on different platforms: games (Wii), control of robotic arms ... As a first step, a study of current gesture recognition techniques that use accelerometers and sensor measurement is performed. In a second step, we study how accelerometers can be used to try to recognize the gestures that can make a person (moving the arm to the side, rotate the hand, draw a square, etc...) And the problems of its use can be derived. We used an IMU (Inertial Measurement Unit) as a measuring sensor. It comprises three accelerometers and three gyroscopes (Xsens MTI-300). The measures provided by these sensors to calculate the position and orientation of the hand are made, with the latter depending on the Euler angles. An important aspect to note is the effect of gravity on the measurements of the accelerations. Through various calculations and with the help of the gyroscopes can correct this effect. Finally, a system that identifies the position and orientation of the hand as recognized gestures developed using fuzzy logic. Both the acquisition of samples to calculate position, a code was developed with Matlab program. Also, with the same software, has implemented a fuzzy logic system to be held with the recognition of gestures using the FIS Editor. Tests have involved the execution of nine gestures by different people having a recognition rate between 90% and 100% depending on the gesture to identify.

Non-invasive user monitoring in daily tasks for cognitive rehabilitation

El uso de técnicas para la monitorización del movimiento humano generalmente permite a los investigadores analizar la cinemática y especialmente las capacidades motoras en aquellas actividades de la vida cotidiana que persiguen un objetivo concreto como pueden ser la preparación de bebidas y comida, e incluso en tareas de aseo. Adicionalmente, la evaluación del movimiento y el comportamiento humanos en el campo de la rehabilitación cognitiva es esencial para profundizar en las dificultades que algunas personas encuentran en la ejecución de actividades diarias después de accidentes cerebro-vasculares. Estas dificultades están principalmente asociadas a la realización de pasos secuenciales y al reconocimiento del uso de herramientas y objetos. La interpretación de los datos sobre la actitud de este tipo de pacientes para reconocer y determinar el nivel de éxito en la ejecución de las acciones, y para ampliar el conocimiento en las enfermedades cerebrales, sus consecuencias y severidad, depende totalmente de los dispositivos usados para la captura de esos datos y de la calidad de los mismos. Más aún, existe una necesidad real de mejorar las técnicas actuales de rehabilitación cognitiva contribuyendo al diseño de sistemas automáticos para crear una especie de terapeuta virtual que asegure una vida más independiente de estos pacientes y reduzca la carga de trabajo de los terapeutas. Con este objetivo, el uso de sensores y dispositivos para obtener datos en tiempo real de la ejecución y estado de la tarea de rehabilitación es esencial para también contribuir al diseño y entrenamiento de futuros algoritmos que pudieran reconocer errores automáticamente para informar al paciente acerca de ellos mediante distintos tipos de pistas como pueden ser imágenes, mensajes auditivos o incluso videos. La tecnología y soluciones existentes en este campo no ofrecen una manera totalmente robusta y efectiva para obtener datos en tiempo real, por un lado, porque pueden influir en el movimiento del propio paciente en caso de las plataformas basadas en el uso de marcadores que necesitan sensores pegados en la piel; y por otro lado, debido a la complejidad o alto coste de implantación lo que hace difícil pensar en la idea de instalar un sistema en el hospital o incluso en la casa del paciente. Esta tesis presenta la investigación realizada en el campo de la monitorización del movimiento de pacientes para proporcionar un paso adelante en términos de detección, seguimiento y reconocimiento del comportamiento de manos, gestos y cara mediante una manera no invasiva la cual puede mejorar la técnicas actuales de rehabilitación cognitiva para la adquisición en tiempo real de datos sobre el comportamiento del paciente y la ejecución de la tarea. Para entender la importancia del marco de esta tesis, inicialmente se presenta un resumen de las principales enfermedades cognitivas y se introducen las consecuencias que tienen en la ejecución de tareas de la vida diaria. Más aún, se investiga sobre las metodologías actuales de rehabilitación cognitiva. Teniendo en cuenta que las manos son la principal parte del cuerpo para la ejecución de tareas manuales de la vida cotidiana, también se resumen las tecnologías existentes para la captura de movimiento de manos. Una de las principales contribuciones de esta tesis está relacionada con el diseño y evaluación de una solución no invasiva para detectar y seguir las manos durante la ejecución de tareas manuales de la vida cotidiana que a su vez involucran la manipulación de objetos. Esta solución la cual no necesita marcadores adicionales y está basada en una cámara de profundidad de bajo coste, es robusta, precisa y fácil de instalar. Otra contribución presentada se centra en el reconocimiento de gestos para detectar el agarre de objetos basado en un sensor infrarrojo de última generación, y también complementado con una cámara de profundidad. Esta nueva técnica, y también no invasiva, sincroniza ambos sensores para seguir objetos específicos además de reconocer eventos concretos relacionados con tareas de aseo. Más aún, se realiza una evaluación preliminar del reconocimiento de expresiones faciales para analizar si es adecuado para el reconocimiento del estado de ánimo durante la tarea. Por su parte, todos los componentes y algoritmos desarrollados son integrados en un prototipo simple para ser usado como plataforma de monitorización. Se realiza una evaluación técnica del funcionamiento de cada dispositivo para analizar si es adecuada para adquirir datos en tiempo real durante la ejecución de tareas cotidianas reales. Finalmente, se estudia la interacción con pacientes reales para obtener información del nivel de usabilidad del prototipo. Dicha información es esencial y útil para considerar una rehabilitación cognitiva basada en la idea de instalación del sistema en la propia casa del paciente al igual que en el hospital correspondiente. ABSTRACT The use of human motion monitoring techniques usually let researchers to analyse kinematics, especially in motor strategies for goal-oriented activities of daily living, such as the preparation of drinks and food, and even grooming tasks. Additionally, the evaluation of human movements and behaviour in the field of cognitive rehabilitation is essential to deep into the difficulties some people find in common activities after stroke. This difficulties are mainly associated with sequence actions and the recognition of tools usage. The interpretation of attitude data of this kind of patients in order to recognize and determine the level of success of the execution of actions, and to broaden the knowledge in brain diseases, consequences and severity, depends totally on the devices used for the capture of that data and the quality of it. Moreover, there is a real need of improving the current cognitive rehabilitation techniques by contributing to the design of automatic systems to create a kind of virtual therapist for the improvement of the independent life of these stroke patients and to reduce the workload of the occupational therapists currently in charge of them. For this purpose, the use of sensors and devices to obtain real time data of the execution and state of the rehabilitation task is essential to also contribute to the design and training of future smart algorithms which may recognise errors to automatically provide multimodal feedback through different types of cues such as still images, auditory messages or even videos. The technology and solutions currently adopted in the field don't offer a totally robust and effective way for obtaining real time data, on the one hand, because they may influence the patient's movement in case of marker-based platforms which need sensors attached to the skin; and on the other hand, because of the complexity or high cost of implementation, which make difficult the idea of installing a system at the hospital or even patient's home. This thesis presents the research done in the field of user monitoring to provide a step forward in terms of detection, tracking and recognition of hand movements, gestures and face via a non-invasive way which could improve current techniques for cognitive rehabilitation for real time data acquisition of patient's behaviour and execution of the task. In order to understand the importance of the scope of the thesis, initially, a summary of the main cognitive diseases that require for rehabilitation and an introduction of the consequences on the execution of daily tasks are presented. Moreover, research is done about the actual methodology to provide cognitive rehabilitation. Considering that the main body members involved in the completion of a handmade daily task are the hands, the current technologies for human hands movements capture are also highlighted. One of the main contributions of this thesis is related to the design and evaluation of a non-invasive approach to detect and track user's hands during the execution of handmade activities of daily living which involve the manipulation of objects. This approach does not need the inclusion of any additional markers. In addition, it is only based on a low-cost depth camera, it is robust, accurate and easy to install. Another contribution presented is focused on the hand gesture recognition for detecting object grasping based on a brand new infrared sensor, and also complemented with a depth camera. This new, and also non-invasive, solution which synchronizes both sensors to track specific tools as well as recognize specific events related to grooming is evaluated. Moreover, a preliminary assessment of the recognition of facial expressions is carried out to analyse if it is adequate for recognizing mood during the execution of task. Meanwhile, all the corresponding hardware and software developed are integrated in a simple prototype with the purpose of being used as a platform for monitoring the execution of the rehabilitation task. Technical evaluation of the performance of each device is carried out in order to analyze its suitability to acquire real time data during the execution of real daily tasks. Finally, a kind of healthcare evaluation is also presented to obtain feedback about the usability of the system proposed paying special attention to the interaction with real users and stroke patients. This feedback is quite useful to consider the idea of a home-based cognitive rehabilitation as well as a possible hospital installation of the prototype.

A combined probabilistic framework for learning gestures and actions

In this paper we introduce a probabilistic approach to support visual supervision and gesture recognition. Task knowledge is both of geometric and visual nature and it is encoded in parametric eigenspaces. Learning processes for compute modal subspaces (eigenspaces) are the core of tracking and recognition of gestures and tasks. We describe the overall architecture of the system and detail learning processes and gesture design. Finally we show experimental results of tracking and recognition in block-world like assembling tasks and in general human gestures.

3D visual sensing of human hand for remote operation of a robotic hand

New low cost sensors and open free libraries for 3D image processing are making important advances in robot vision applications possible, such as three-dimensional object recognition, semantic mapping, navigation and localization of robots, human detection and/or gesture recognition for human-machine interaction. In this paper, a novel method for recognizing and tracking the fingers of a human hand is presented. This method is based on point clouds from range images captured by a RGBD sensor. It works in real time and it does not require visual marks, camera calibration or previous knowledge of the environment. Moreover, it works successfully even when multiple objects appear in the scene or when the ambient light is changed. Furthermore, this method was designed to develop a human interface to control domestic or industrial devices, remotely. In this paper, the method was tested by operating a robotic hand. Firstly, the human hand was recognized and the fingers were detected. Secondly, the movement of the fingers was analysed and mapped to be imitated by a robotic hand.

Multimodal 'Eyes-Free' interaction techniques for mobile devices

Mobile and wearable computers present input/output prob-lems due to limited screen space and interaction techniques. When mobile, users typically focus their visual attention on navigating their environment - making visually demanding interface designs hard to operate. This paper presents two multimodal interaction techniques designed to overcome these problems and allow truly mobile, 'eyes-free' device use. The first is a 3D audio radial pie menu that uses head gestures for selecting items. An evaluation of a range of different audio designs showed that egocentric sounds re-duced task completion time, perceived annoyance, and al-lowed users to walk closer to their preferred walking speed. The second is a sonically enhanced 2D gesture recognition system for use on a belt-mounted PDA. An evaluation of the system with and without audio feedback showed users' ges-tures were more accurate when dynamically guided by au-dio-feedback. These novel interaction techniques demon-strate effective alternatives to visual-centric interface de-signs on mobile devices.

Analysis and Evaluation of Tiny Machine Learning applications

The aim of TinyML is to bring the capability of Machine Learning to ultra-low-power devices, typically under a milliwatt, and with this it breaks the traditional power barrier that prevents the widely distributed machine intelligence. TinyML allows greater reactivity and privacy by conducting inference on the computer and near-sensor while avoiding the energy cost associated with wireless communication, which is far higher at this scale than that of computing. In addition, TinyML’s efficiency makes a class of smart, battery-powered, always-on applications that can revolutionize the collection and processing of data in real time. This emerging field, which is the end of a lot of innovation, is ready to speed up its growth in the coming years. In this thesis, we deploy three model on a microcontroller. For the model, datasets are retrieved from an online repository and are preprocessed as per our requirement. The model is then trained on the split of preprocessed data at its best to get the most accuracy out of it. Later the trained model is converted to C language to make it possible to deploy on the microcontroller. Finally, we take step towards incorporating the model into the microcontroller by implementing and evaluating an interface for the user to utilize the microcontroller’s sensors. In our thesis, we will have 4 chapters. The first will give us an introduction of TinyML. The second chapter will help setup the TinyML Environment. The third chapter will be about a major use of TinyML in Wake Word Detection. The final chapter will deal with Gesture Recognition in TinyML.

Optimizing AI at the Edge: from network topology design to MCU deployment

The first topic analyzed in the thesis will be Neural Architecture Search (NAS). I will focus on two different tools that I developed, one to optimize the architecture of Temporal Convolutional Networks (TCNs), a convolutional model for time-series processing that has recently emerged, and one to optimize the data precision of tensors inside CNNs. The first NAS proposed explicitly targets the optimization of the most peculiar architectural parameters of TCNs, namely dilation, receptive field, and the number of features in each layer. Note that this is the first NAS that explicitly targets these networks. The second NAS proposed instead focuses on finding the most efficient data format for a target CNN, with the granularity of the layer filter. Note that applying these two NASes in sequence allows an "application designer" to minimize the structure of the neural network employed, minimizing the number of operations or the memory usage of the network. After that, the second topic described is the optimization of neural network deployment on edge devices. Importantly, exploiting edge platforms' scarce resources is critical for NN efficient execution on MCUs. To do so, I will introduce DORY (Deployment Oriented to memoRY) -- an automatic tool to deploy CNNs on low-cost MCUs. DORY, in different steps, can manage different levels of memory inside the MCU automatically, offload the computation workload (i.e., the different layers of a neural network) to dedicated hardware accelerators, and automatically generates ANSI C code that orchestrates off- and on-chip transfers with the computation phases. On top of this, I will introduce two optimized computation libraries that DORY can exploit to deploy TCNs and Transformers on edge efficiently. I conclude the thesis with two different applications on bio-signal analysis, i.e., heart rate tracking and sEMG-based gesture recognition.

Hand gesture spotting and recognition using HMMs and CRFs in color image sequences

Magdeburg, Univ., Fak. für Elektrotechnik und Informationstechnik, Diss., 2010

A SDK improvement towards gesture support

Human-Computer Interaction have been one of the main focus of the technological community, specially the Natural User Interfaces (NUI) field of research as, since the launch of the Kinect Sensor, the goal to achieve fully natural interfaces just got a lot closer to reality. Taking advantage of this conditions the following research work proposes to compute the hand skeleton in order to recognize Sign Language Shapes. The proposed solution uses the Kinect Sensor to achieve a good segmentation and image analysis algorithms to extend the skeleton from the extraction of high-level features. In order to recognize complex hand shapes the current research work proposes the redefinition of the hand contour making it immutable to translation, rotation and scaling operations, and a set of tools to achieve a good recognition. The validation of the proposed solution extended the Kinects Software Development Kit to allow the developer to access the new set of inferred points and created a template-matching based platform that uses the contour to define the hand shape, this prototype was tested in a set of predefined conditions and showed to have a good success ration and has proven to be eligible for real-time scenarios.

Recognition of Facial Expressions with Autoencoders and Convolutional-Nets

Les humains communiquent via différents types de canaux: les mots, la voix, les gestes du corps, des émotions, etc. Pour cette raison, un ordinateur doit percevoir ces divers canaux de communication pour pouvoir interagir intelligemment avec les humains, par exemple en faisant usage de microphones et de webcams. Dans cette thèse, nous nous intéressons à déterminer les émotions humaines à partir d’images ou de vidéo de visages afin d’ensuite utiliser ces informations dans différents domaines d’applications. Ce mémoire débute par une brève introduction à l'apprentissage machine en s’attardant aux modèles et algorithmes que nous avons utilisés tels que les perceptrons multicouches, réseaux de neurones à convolution et autoencodeurs. Elle présente ensuite les résultats de l'application de ces modèles sur plusieurs ensembles de données d'expressions et émotions faciales. Nous nous concentrons sur l'étude des différents types d’autoencodeurs (autoencodeur débruitant, autoencodeur contractant, etc) afin de révéler certaines de leurs limitations, comme la possibilité d'obtenir de la coadaptation entre les filtres ou encore d’obtenir une courbe spectrale trop lisse, et étudions de nouvelles idées pour répondre à ces problèmes. Nous proposons également une nouvelle approche pour surmonter une limite des autoencodeurs traditionnellement entrainés de façon purement non-supervisée, c'est-à-dire sans utiliser aucune connaissance de la tâche que nous voulons finalement résoudre (comme la prévision des étiquettes de classe) en développant un nouveau critère d'apprentissage semi-supervisé qui exploite un faible nombre de données étiquetées en combinaison avec une grande quantité de données non-étiquetées afin d'apprendre une représentation adaptée à la tâche de classification, et d'obtenir une meilleure performance de classification. Finalement, nous décrivons le fonctionnement général de notre système de détection d'émotions et proposons de nouvelles idées pouvant mener à de futurs travaux.