In this paper we present an approach aimed at optimizing communications in a distributed smart camera network. The goal is to further reduce the amount of video data to be compressed and transported across the network, thus reducing processing time on smart camera nodes and increasing the network bandwidth. We seek to transmit only data of major importance in region of interest of captured frames. These data are then compressed and transmitted across the network. The integration of this data reduction scheme in a fast communication and object exchange mechanism among smart camera nodes allows for a real-time processing and coordination of distributed video applications. Using a hardware/software architecture on FPGA-based smart cameras, we further improve the computational and communication performance of the whole system, with time critical parts mapped in hardware while control dominated and high-level modules are maintained in software. Results show that the payload obtained after compressing just the ROI parts of a frame is considerably small to allow every camera node to be able to process about 30 frames and transmit up to 1500 compressed frames per second.