Skip to Main Content
This paper presents a new approach to data fusion for automatic recognition, surveillance, and tracking in intelligent transportation systems. Robust data alignment (RDA), i.e., finding relational maps among a sequence of invariant feature data sets, is one of the key requirements for successful data fusion. To achieve RDA for correspondenceless data fusion, we construct a cost criterion based on the information theory and solve an optimization problem with a mixed search strategy that combines the Nelder-Mead simplex and random search methods. We evaluate the cost criterion and search strategy by a numerical stability test and suggest an outlier rejection technique for refining the previous feature data and, at the same time, extracting moving vehicles that are contained in the collected outliers. Experimental results on a video sequence that is collected from an unmanned aerial vehicle indicate the potential of aerial monitoring and tracking systems built on our information-theoretic RDA.