Road and Scene Recognition for the Analysis of Egocentric…
Abstract
Road recognition for safe driving purposes has been a widely investigated subject in computer vision for over forty years. In the literature road recognition refers to having camera(s) fixed inside of a vehicle, which is four-wheeled, looking along the movement direction. In some cases some other active or passive sensing devices have been used for better understanding of the three dimensional scene, hence, escalating the robustness. This thesis aims to introduce a real ego-centric analysis of the road and its surrounding scene from a single point camera attached to motorcyclist’s helmet or wore by the motorcyclist. One reason for choosing motorcyclists is the large number of publicly available videos. Different from the usual approaches, instead of having multiple sensors for leveraging the level of information, combining scene recognition road recognition on semantic basis would be considered as the second major aim of this thesis. Both of these aims are also the novelties of this thesis as they have not been investigated together in this field yet. Above these two aims, on humanitarian level this thesis expects to take a pace towards safe driving assistance for cyclists and motorcyclist which has not been considered in the literature or in the industry.