Riconoscimento Visuale di Forme e Luoghi su Dispositivi Mobili

Abstract: The growth of mobile devices capabilities makes them suitable to perform complex processing tasks. This greatly widens the range of algorithms that can be run directly on the mobile device, therefore enabling the spread of many new applications unfeasible until few years ago. Researches in Computer Vision can now exploit the growing computational capabilities of mobile devices equipped with high quality cameras, as long as many other built-in sensors. However, several limitations of mobile device, with respect to traditional desktop computers, must take into account. Despite the hardware improvements, computational capabilities and memory availability may present a severe issue, as well as limited battery life and network connectivity. Also, in the mobile context the user directly interacts with the device so that real time response is often required. Such limitations suggest that moving the computation towards mobile devices is not a mere porting of existing algorithms. Optimized code may run on the device, but most application require further processing or data that cannot be found directly on the mobile device. Building mobile application requires to design algorithms that fit in the system architecture composed in by the mobile device itself, a remote server and the network connectivity in between. The purpose of the recently born field of Mobile Vision is to face these issues. Mobile Vision is not only about optimizing computer vision algorithms to run on limited hardware, but also about defining mobile-oriented paradigms for algorithms, and application designs to meet a particular mobile vision system architecture, exploiting the set of sensors available on the mobile device, and taking advantage of the role played by the user in a mobile context. The goal of this thesis is twofold. Firstly, it explores the improvements brought so far thanks to Mobile Vision, providing a thorough analysis of the literature in this field and focusing on the open challenges. The architectural solutions and the optimization techniques required to run mobile vision applications are then discussed. Secondly, it proposes two novel applications, namely an algorithm that make the ellipse detection task feasible on mobile device in real-time, and a lightweight approach to visual place recognition to provide on the fly useful content to users through intuitive and natural interaction.

Citation:

Fornaciari, Michele "Riconoscimento Visuale di Forme e Luoghi su Dispositivi Mobili" 2014

 not available

Paper download:

• Author version: