Traffic video surveillance applications are increasingly maturing to intelligent infrastructures based on networks of embedded smart cameras. These smart cameras provide on-board real-time video analysis and streaming. In previous work a hardware platform for a smart camera was proposed. This work now presents a software framework for power-aware service reconfiguration in embedded multiDSP smart cameras. An efficient publisher-subscriber communication scheme together with dynamic loading capabilities enables reconfiguration at runtime. Depending on the observed scenario an optimal camera configuration is chosen. A multicriterion optimizer implemented as a genetic online algorithm continuously computes optimal algorithm configurations by minimizing power-consumption and maximizing quality-of-service. Based on a cost-function an appropriate configuration is chosen among the set of Pareto-optimal solutions
展开▼