Carnegie Mellon University
Low Power Design Techniques for Video Sensor Networks
Wireless sensor networks (WSN) operating on limited energy resources need to be power efficient to extend system lifetime. This is especially challenging for video sensor networks (VSNs) due to the large volume of data which processing elements (PE) must operate on in short periods of time. For example, a 640x480 video stream recorded at 30 frames per second results in roughly 9 million pixels per second. When compared to temperature sensors (sensing at 1Hz or below), accelerometers (10s of Hz), and other sensing modes, this volume of data means that special consideration needs to be made to bring standard video processing techniques into cost affordable realms so that battery resources for VSNs can last for months or years instead of hours or days.
This talk introduces video processing techniques geared towards achieving three main goals:
- Lower the processing and storage requirements per PE
- Extend the VSN system lifetime
- Shorten the time each node remains active while maintaining acceptable QoS levels for the application
This talk will introduce techniques which are designed to lower power consumption and extend system lifetime of future VSNs. The techniques focus on both node-level design and network-level aspects, as energy-awareness should happen at all levels of abstraction. At node-level, complexity is reduced by processing only a part of each video frame, called region-of-interest (ROI) processing and by distributing the processing tasks to multiple PEs (adaptive data partitioning, or ADP). These techniques help lower processing requirements per PE significantly while maintaining acceptable QoS. At network-level, traditional power management policies are extended to consider other nodes within the network in a distributed fashion. Specifically, distributed power management (DPM) will be proposed and shown to be more effective than the local, well-known counterparts.
A detailed VSN simulator has been developed for evaluating these various power-aware techniques. Further, first generation video sensor network prototype nodes have been built to demonstrate the techniques.
Nick Zamora earned a B.S. degree in Electrical Engineering and Computer Science from U.C. Berkeley in 2001, and an M.S. degree in Electrical and Computer Engineering from CMU in 2003. The contents of this talk constitute much of his thesis topic.