Assia Benbihi presents Autonomous Long-Term Visual Monitoring of Natural Environments

On 2020-01-07 11:00:00 at G205, Karlovo náměstí 13, Praha 2
The goal of visual monitoring is to assess the evolution of an
environment. The nature of these variations is different for natural
scenes than urban ones. For example, tracking the evolution of trees
over the years is different than tracking buildings. In this talk, I
will present my work on autonomous solutions to some of the main
challenges in visual monitoring of natural environments.

In the first part of the talk, I will briefly discuss how to choose
where to look and how to quantify the variations of the scene using
semantics. I will then describe recent work on scene alignment with
local features extracted from a trained CNN. Finally, I will present
current work on scene tracking across seasons using image descriptors
based on the fusion of geometry and semantics.
Responsible person: Petr Pošík