This dissertation investigates energy-saving techniques in battery-powered wireless networks of energy-hungry sensors, such as video cameras and gas sensors for smart surveillance applications. The proposed power management is performed at three levels: sensor, node and network level. The research is performed on commercial off-the-shelf wireless sensor nodes and on laboratory prototypes of wireless sensor nodes. The proposed context-aware multimodal and heterogeneous architecture of the sensor network limits the energy-hungry sensor’s active time and area, while wake-up radio ensures time- and energy-efficient communication between sensor nodes. By combining energy-savings in communication and sensing, we attain runtime prolongation of a wireless sensor node and network, without downgrading quality of service for the user (i.e., maintaining good recognition of interesting events). Moreover, the response latency in our approach is reduced for two orders of magnitude with respect to the state-of-the-art solutions. We show that a holistic approach to design complex surveillance system is necessary to ensure long-living autonomous smart devices. In particular, the scientific contributions of this thesis are: 1) Multilevel approach to enhance energy saving in wireless gas sensor network; 2) Power consumption vs. delay analysis of using wake-up receivers in wireless sensor networks with energy-hungry sensors; 3) Energy-saving method for event detection in wireless video sensor network using heterogeneous sensors and wake-up receivers; 4) Energy-efficient architecture of wireless sensor network with camera and gas sensors for surveillance and safety monitoring systems.