Title: Object matching in large scale visual sensor networks

Abstract:

Visual sensors detect objects and represent them by features of various dimensions which can be distributed to other sensors. Given the newly acquired object features at one of the nodes, the problem is to find whether any of the (distant) nodes in the network has seen same or similar object before, while minimizing communication overhead of combined feature distribution and search for matching. Existing hierarchical feature distribution schemes  flood features with reduced granularity with distance and establishes back pointers, which are used to search for matching of newly detected object features. We propose adaptation of a quorum based scheme  (with rows of sensors for distribution and columns for search), combined with feature reduction. We then generalize feature distribution by constructing information mesh generated by previous occurrences of same object at several sensors in the network. Experiments will be performed using histogram matching for feature comparisons, and publicly available image databases. The main comparison criterion is scalability, protocol performance with increased number of nodes. The next criterion is the actual accuracy of matching outcome in tested applications. Finally, an important goal is to study the impact of various feature hierarchies.