Surveillance Architecture for Human Activity Recognition using Unmanned Aerial Vehicle

Abstract

There is an intensive growth in researches regarding surveillance and threat detection. Surveillance tasks often involve several actors with multiple interactions. Thus, modeling a complex activity becomes challenging. This work proposes an architecture comprised of low, middle, and high levels. The low-level recognizes characteristics, positioning of objects, and time of occurrences utilizing a camera and Unmanned Aerial Vehicle (UAV) sensors. The middle-level is responsible for structuring the information from the low-level using Deterministic Finite Automata (DFA). An expert system attached in the high-level module performs inference over the organized information to enables the system to have simple reasoning modules, assisting the operator decision. The architecture is embedded in a UAV to reduce the number of cameras and to reach difficult areas. The experiments showed that the proposed system updated the grammatical structure effectively given a sequence of information computed by the vision modules.