Above is an overview of the searise target-system, illustrating both algorithmic and hardware modularization. The large boxes represent the three nodes of the computational network: The camera system, the compute server, and the storage server. These systems are connected via fast ethernet and fibre-channel. Computation in the compute server are separated into low-level operations, which are relatively simple algorithmically but computationally expensive due to the amount of visual details that need to be processed, and higher level vision, which is more complex but also processes more abstract and compact data representations. The former is implemented on GPU hardware which is ideal for this kind of simple, high-throughput processing, while the latter is implemented on multiple CPU cores which are more flexible. For details on the work on the various components and their integration follow the links below.
