Performance evaluation of object recognition tasks in visual sensor networks

We analyze the performance of TCP and UDP transport protocols when applied to image retrieval or object recognition in wireless visual sensor networks (VSN). We focus on two different paradigms for image analysis, namely compress-then-analyze (CTA) and analyze-then-compress (ATC). The former entails the transmission of JPEG encoded images from camera nodes to a server, where the analysis takes place. The latter consists in extracting and compressing local visual features on board camera nodes, before transmission to a remote location. The presented analysis is useful to assess the best coupling between application and transport layers under delay and accuracy constraints for different networking conditions.