A multi-perspective aerial monitoring system for scenario detection

Abstract

Detecting alerting or dangerous situations by a Unmanned Aerial Vehicle (UAV) can be a very tricky activity, especially when videos present articulated and dynamic daily life scenarios, with several humans and not human actors triggering lots of events. Furthermore, scenarios are set in different environments and sometimes the scene interpretation be strictly affected by the environment where the scene evolves. To address this issue, this paper presents a novel model to detect real dynamic scenarios occurring in UAV videos, by combining cognitive science methodologies with semantic technologies to build a multi-perspective mental landscape of the scenario. The semantic, ontology-based description of the scenario provides the spatio/temporal context of the current scene, and supports the objects detection and their main interactions. The spatio/temporal context allows the automatic generation of a multi-perspective Fuzzy Cognitive Map (FCM), which is built by merging several FCMs on single scenario objects, their interactions and general aspects of the scenario. The FCM provides evaluations of possible scenario evolutions. A case study and preliminary test results show the applicability of the proposed model to alerting scenario detection.