This paper proposes an approach to the fusion of multi-modal sensor data for the purpose of personnel intrusion detection.
The focus is on using low cost non-imaging sensors for applications such as border crossings where issues of rapid deployment and power consumption are prevalent.
The main challenge of fusing data from such sensors lies in the wide variation of granularity of classification that they may provide.
While some sensors may provide detailed characteristics of the motion in a scene and therefore a very fine classification,others may only provide simple alerts and little detail. In order
to fuse data from a wide range of sensors that are often designed for disparate applications, an approach based on the Dempster-Shafer theory of evidence is used. The implicit handling of uncertainty and ambiguous propositions leads to a convenient hierarchical approach that can represent data from numerous sensor modalities