Optimal object association from pairwise evidential mass functions

Object association is often a prior step in the data fusion process, especially for multiple objects tracking and multisensor data fusion. The approach introduced in this paper associates objects detected in a scene by two sensors, while modeling uncertainty using the Dempster-Shafer theory of belief functions. Sensor information is transformed into pairwise mass functions, which are combined using Dempster's rule of combination. The result of this combination allows us to find the most plausible relation between two sets of objects by solving a linear programming problem. Experimental results with real data acquired from sensors embedded in intelligent vehicles are presented.

I. Introduction

Object association is often a prior step in the data fusion process. This paper deals with the association of objects detected by two sensors in their area of interest. Sensors treat observations and provide information about objects such as number of detected objects, kinematic information (position, velocity), identification information (status, class, dimensions, etc.). Associating objects is a difficult problem since the number of objects in the scene is usually unknown and data provided by sensors can be uncertain and incomplete (with possible false alarms and non detections).