Navigational strategies and surveillance

Abstract

This paper will describe new work that attempts to perform the modelling of human behaviour not at the level of visible patterns of motion, but at the level of intentions. By inferring intentions in terms of known goals, it becomes possible to explain the behaviour of people moving around within the field of view of a video camera (e.g. ”Agent 25 went to exit 8 via sub-goals 34 and 21”). Earlier work used an adhoc model of human navigation and recalculated possible intentions at each frame, whereas the work presented here incorporates models of navigation from within psychology which are both simpler and more conceptually plausible, whilst providing comparable results. The basic algorithm involves generating all possible plausible paths through the scene to known goal sites, and then measuring the distance between each path and the agent’s actual trajectory. Two navigational strategies are discussed, and a number of distance measures are proposed and evaluated. A prototype system has been tested on video from an outdoor car-park and an indoor foyer scene, and it has been found to produce psychologically plausible explanations in the majority of cases.