In-Network Processing
We are developing methods for processing complex data such as that from audio and visual sensors within a wireless sensor network where there are severe resource constraints in terms of processing power, bandwidth, and available energy resources. The ability to handle complex data will increase the utility of wireless sensor networks beyond those applications for which simple scalar data (such as temperature or humidity measurements) is sufficient. Current deployments utilising our research outcomes include installations of acoustic sensors for monitoring bird and frog populations.
Research Topics
Efficient Processing of Audio/Video Information
The recent rapid drop in the physical size, cost and energy requirements of CMOS cameras, DSP coprocessors and audio codecs, provides an opportunity to collect audio and video/image data using networks of sensors. Although this provides a rich, complex source of data for extaction of information, it also presents some significant scientific and engineering challenges in finding ways to efficiently process this type of data. Our research in this area is focussed on methods for efficient filtering and feature extraction from multi-modal data.
Distributed decision making
One of the key advantages that can be gained from using sensor networks, over more traditional means of gathering data, is to make use of the spatial distribution of points of measurement. Rather than naively processing data at all nodes in an independent fashion, the ability for nodes to communicate with each other can be exploited considerably. For instance, where a number of nodes are concurrently taking observations of a single event, nodes can gradually build up confidence about the quality of their own obervations. By sharing this confidence level around, nodes with low confidence (e.g. far away from the event) can cease their own processing to save on energy, knowing that other nodes in the network will capture the event. This kind of co-operative approach to event processing, can transform the kind of information sensor networks can return about the environment.
In-network storage of complex data
An ongoing problem for sensor networks is dealing with often low-quality and dynamic radio links. This is a particularily important problem when dealing with more complex data types as this may often require a very large number of consecutive packets to all be successfully delivered to a gateway in order for sensible interpretation of events to be possible (e.g. compressed image or audio segment). We are researching methods for in-network storage where multiple packets from a single group of information may be stored all over the network depending on the current quality of links and status of connectivity. We are investigating increasingly smart methods for regrouping of this data to ensure a good balance between both latency and fidelity of information being returned.
Efficient compression techniques
Radio usage is one of the biggest consumers of energy is a sensor network. As such the goal of any network is to minimise the amount of data which needs to be passed between nodes. In cases where a representation of the raw sample data is required at the base, compression of data is a must. Our research is addressing the challenges of dealing with compression of complex data in sensor networks, including the efficiency of compression algorithms and the ability of a chosen codec to deal with missing data.
In-network interpolation / data representation
The end-application will often define the desired minimum sample rate for a particular phenomena being measured. Given unlimited energy, bandwidth and processing resources then this requirement can always be satisfied. However, in case where any of these resources are limited, it may not always be possible to deliver samples at the rate specified. We are resarching alternative means for dealing with data samples under these types of conditions.
Key Staff
Publications
Recent publications in the area of in-network processing of complex data in sensor networks:
- Wen Hu, Nirupama Bulusu, Chun Tung Chou, Andrew Taylor, Van Nghia Tran, Sanjay Jha. "The design and evaluation of a hybrid sensor network for cane-toad monitoring." ACM Trans. Sensor Networks. 2009; 5(1):Art. 4 (28 pp.).
- Darren Moore. "Demonstration of bird species detection using an acoustic wireless sensor network." 33rd IEEE International Conference on Local Computer Networks (LCN 2008) : SenseApp 2008; Montreal, Que. IEEE; 2008: 730-731.
- Thanh Dang, Nirupama Bulusu, Wen Hu. "Lightweight Acoustic Classification For Cane-Toad Monitoring." Proceedings of the Asilomar Conference on Signals, Systems and Computers, Pacific Grove, California, October 2008.
- Tim Wark, Peter Corke, Jim Liu, Darren Moore. "Design and evaluation of an image analysis platform for low-power, low-bandwidth camera networks." Workshop on Applications, Systems, and Algorithms for Image Sensing (ImageSense '08); Raleigh, NC. ACM; 2008: 7-11.
