Skip to Main content Skip to Navigation
Theses

Conception et réalisation d'un système multi-fonctionnel d'aide à la mobilité pour personnes mavoyantes et aveugles.

Abstract : Visual impairment and blindness are sources of mobility difficulties for the affected people. In orderto lighten the burden of these difficulties, many mobility aids have been imagined, designed, tested, and more or less adopted. Designers of such assistive systems soon run into the complexity of the issue, which stands at the intersection of three domains that are, by themselves, complex: visual impairment, mobility, and perception.Having decided to design an electronic assistive system from the beginning, we tried to step back and analyzed a wide range of blind aids: white canes, mobile electronic devices serving different mobility purposes, urban systems, and systems not primarily designed for mobility. Their diversity helped us analyze assistive systems through several prisms: according to their technical characteristics, their functions, their shape, and their dependence to an infrastructure. Taken individually, each of these approaches quickly shows its limits, but, together, they draw an interesting portrait of the existing devices. Besides these rather classical approaches, we propose a new model for analyzing assistive systems, which relies on the way these systems take place in a person's perception / mobility process. This model has the advantages of being, a priori, relevant forall assistive systems – in spite of their dissimilarity – and meaningful for both evaluation and classification.We have designed and built an electronic mobility aid, called the 2SEES system. Like its predecessor, the SEES system, the 2SEES relies on three platforms: a smart cane, a smartphone, anda cloud computing back-end. The issues of energy consumption and geographical universality, essential for any mobile device, are made explicit and studied. A novelty of the 2SEES resides in its account of reliability issues ; it is thus designed around the complex equilibrium between energy consumption, universality, and reliability. Despite their importance in a system destined to be adopted by end users, these three notions are scarcely visible in the relevant literature.To concurrently enhance robustness and autonomy, we have integrated several sensors and processors in the smart cane, both by introducing redundancy, for fault tolerance, and by integratingheterogeneous sensors, for robustness against the diversity of environments.Two aspects of this equilibrium have been further studied. First, the need for robustness has been highlighted by a study of affinities between obstacle sensors and several types of potential obstacle materials. Secondly, we have tried to develop an energy-efficient indoor localization function that islittle dependent on infrastructures, and therefore easily scalable. This function works with embedded sensors (wheel encoder, inertial measurement unit) and a simplified particle filter, which estimates the position by checking the coherence of trajectories derived from sensor data against themap of the location.In addition to this work on the balance between robustness, energy consumption, and universality, we have developed a novel function, named SO2SEES, which allows communication between usersof the 2SEES and smart objects. This function enables users to ask, in natural language, questions tothe 2SEES, which are answered using information coming from surrounding smart objects. In orderto keep the system simple, users do not formulate their own questions, but are instead invited to pick them in a set of predefined questions, which are proposed by the system according to the nearby objects and the information they offer. This mode of operation shifts the system from a natural language processing artificial intelligence to an expert system working on dynamic and distributed knowledge bases. In this latter configuration, the main issue is the interoperability between the 2SEES and the smart objects and their back-ends that take part in the functionality.
Complete list of metadatas

Cited literature [197 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-02926122
Contributor : Abes Star :  Contact
Submitted on : Monday, August 31, 2020 - 1:56:12 PM
Last modification on : Friday, September 18, 2020 - 4:06:14 AM

File

2019CLFAC097_CONNIER.pdf
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-02926122, version 1

Collections

Citation

Jean Connier. Conception et réalisation d'un système multi-fonctionnel d'aide à la mobilité pour personnes mavoyantes et aveugles.. Intelligence artificielle [cs.AI]. Université Clermont Auvergne, 2019. Français. ⟨NNT : 2019CLFAC097⟩. ⟨tel-02926122⟩

Share

Metrics

Record views

82

Files downloads

20