Self-Organizing, Adaptive Data Fusion for 3d Object Tracking

Data fusion concepts are a necessary basis for utilizing complex networks of sensors. A key feature for a robust data fusion system is adaptivity, both to be fault-tolerant and to run in a self-organizing manner. In this contribution a general framework for adaptive data fusion is established with object tracking as an application. The fusion algorithm of Democratic Integration is presented as one possible robust approach to the fusion task. As an alternative the STAPLE algorithm will be shown, which was previously only used for late classifier fusion. Extensions to apply the STAPLE algorithm for the fusion of probabilities will be introduced. Finally both algorithms will be evaluated on complex, realistic scenes to show their capabilities of self-organization and fault-tolerance.