Multiple-View Object Recognition in Smart Camera Networks

We study object recognition in low-power, low-bandwidth smart camera networks. The ability to perform robust object recognition is crucial for applications such as visual surveillance to track and identify objects of interest, and overcome visual nuisances such as occlusion and pose variations between multiple camera views. To accommodate limited bandwidth between the cameras and the base-station computer, the method utilizes the available computational power on the smart sensors to locally extract SIFT-type image features to represent individual camera views. We show that between a network of cameras, high-dimensional SIFT histograms exhibit a joint sparse pattern corresponding to a set of shared features in 3-D. Such joint sparse patterns can be explicitly exploited to encode the distributed signal via random projections. At the network station, multiple decoding schemes are studied to simultaneously recover the multiple-view object features based on a distributed compressive sensing theory. The system has been implemented on the Berkeley CITRIC smart camera platform. The efficacy of the algorithm is validated through extensive simulation and experiment.