| Summary: | In recent years, there has been interest in detailed monitoring of road traffic, particularly in intersections, in order to obtain a statistical model of the flow of vehicles through them. While conventional methods - sensors at each of the intersection's entrances/exits - allow for counting, they are limited in the sense that it is impossible to track a vehicle from origin to destination. This data is invaluable to understand the how the dynamic of a city's mobility works, and how it can be improved, therefore new techniques must be developed which provide that kind of information. One of the possible approaches to this problem is to analyse video footage of said intersections by means of computer vision algorithms, in order to identify and track individual vehicles. One of the possible ways to obtain this footage is by flying a drone - a small unmanned air vehicle (UAV) - carrying a camera over an intersection.Some work has been done with this solution in mind, but the usage of a top-down birds-eye perspective, obtained by flying the drone directly above the intersection, rather than at an angle, is limited or inexistent. This approach is interesting because it circumvents the problem of occlusions present in other footage capture set ups. The focus of this dissertation is, then, to develop and apply computer vision algorithms to footage obtained in this way in order to identify and track vehicles across intersections, so that a statistical model may be extracted. This model is based on said association of an origin and a destination. Based on the implementation which was developed, this approach seems to be useful for at least some types of vehicles.
|
|---|