Multi-Object Association with Belief Functions: the Dual Decision Algorithm

In intelligent vehicles applications, Multi-Target Tracking (MTT) is usually done considering imperfect sensors' data. As belief functions and particularly the Transferable Belief Model (TBM) help to model these imperfections, it represents an interesting framework for MTT. To carry out target tracking, the system must be able to define at at time k the relations between the detected objects (targets) and the already known ones (tracks). The correlation step is called Multi-Object Association (MOA). This paper tackles the problem of TBM-MOA decision making. The aim is to present a methodology and measures for the management of associations in presence of conflicting sensors' measures. A new algorithm based on a single association matrix gathering all the knowledge of the targets tracks associations is presented. It provides a semantic to avoid the aforementioned difficulties. Simulation results of an obstacle detection application show the advantages of the proposed solution.