Unified situation modeling and understanding using hierarchical graphical Model

Abstract

Situation modeling and understanding is a challenging task due to the scene complexity. In this work we use the extended Joint Directors of Laboratories model for sensor data fusion to solve this challenging task. We propose a non-directed graphical model to represent the global and local contextual dependencies (spatial, temporal, and semantic) between the objects (pedestrians, vehicles, and lanes) and the scene. We develop an inference algorithm to estimate the probability density function at each node of the graph in a bottom up top down approach using non-parametric belief propagation (NBP) scheme. The inferred objects are contextually consistent with respect to others objects and the scene.