Human Action Recognition in Surveillance Videos using Abductive Reasoning on Linear Temporal Logic
Saikat Kumar Basu, Malcolm Stagg, Robert DiBiano, Manohar Karki, Supratik Mukhopadhyay · 2012
Abstract : Real time motion tracking is a very important part of activity recognition from streaming videos. But little research has been done in recognizing the top-level plans linking the atomic activities evident in various surveillance footages. This paper proposes a novel approach for high-level action recognition in surveillance videos combining Linear Temporal Logic (LTL) and Abductive Reasoning. Although both LTL and Abductive reasoning have been used separately for plan recognition in various Artificial Intelligence (AI) systems and mobile robots, the framework proposed in this paper combines the two by first mapping the surveillance videos to LTL formula and then using probabilistic and logical reasoning to identify complex events like burglary/escapade or deal with arbitrary events like occlusion or random stops.