Motivation

Smart Homes for the aged:

This project aims to provide operational effectiveness for people with decreasing functional capacity. In particular, we tackle the scientific and technical issues in the context of building smart spaces for the aged and for people with varying and decreasing abilities. The overall goal is to produce a system that will monitor and support activities of varying complexity without compromising a normal lifestyle. Building such a system requires the learning and recognition of behaviors and activities of people from multi-sensor data as they go about their daily schedules.

Wide area surveillance systems:

This includes the development of tools and technologies for large scale surveillance both with camera and other sensory data. This differs from the Smart Home project because we are dealing with large number of people (known and unknown) over much bigger areas (floors, buildings etc.) as well as certain people having intentional suspicious activity.

Published Papers

1) Nguyen, N.T., Venkatesh, S., West, G.A.W. & Bui, H.H., (2003), Multiple camera coodination in a surveillance system, Acta Automatica Sinica, Vol. 29, No. 3, May, pp. 408-422.

2) Bui, H. H., Venkatesh, S. & West, G. A. W., (2001), Policy recognition in the abstract Markov model, Journal of Artificial Intelligence Research, Vol. 17, pp. 451-499.

3) Bui, H. H., Venkatesh, S. & West, G. A. W., Layered dynamic probabilistic networks for spatio-temporal modelling, Intelligent Data Analysis, vol 3, no 5, pp 339-361, 1999.

4) Bui, H. H., Venkatesh, S. & West, G. A. W., (2001), Tracking and surveillance in wide-area spatial environments using the abstract hidden Markov model, Invited paper for the Special Issue of the International Journal on Pattern Recognition and Artificial Intelligence on "Hidden Markov Models in Vision", Vol. 15, No. 1, February.

5) Luhr, S., Bui, H. H., Venkatesh, S. & West, G. A. W., (2003), Recognition of human activity through hierarchical stochastic learning, Proc. IEEE International Conference on Pervasive Computing, Texas, March.

6) Peursum, P., Venkatesh, S., West, G. A. W. & Bui, H. H., (2003), Object labelling from human action recognition, Proc. IEEE International Conference on Pervasive Computing, Texas, March.

7) Nguyen, N., Venkatesh, S., West, G. A. W. & Bui, H. H., (2002), Hierarchical monitoring of peoples behaviour in complex environments using multiple cameras, Proc. International Conference on Pattern Recognition, Quebec, August 11-15.

8) Nguyen, N., West, G. A. W., & Venkatesh, S., (2002), Coordination of multiple cameras to track people, Asian Conference on Computer Vision, Melbourne, January 23-25.

9) Nguyen, N.T., Bui, H.H., Venkatesh, S. & West, G.A.W., (2003), Recognising and Monitoring High-Level Behaviours in Complex Spatial Environments, Proc. Conf. Computer Vision and Pattern Recognition, Madison, Wisconsin, USA, June.

Funding

The above two projects are funded through an ARC Centre of Excellence and ARC Discovery grants.

1. Challenges in Surveillance/Monitoring and Pattern Recognition/Machine Learning
2. With the advent of multiple, distributed multi-modal sensors, the key issues are:
3. The development of representational frameworks that can model activities and the sensory data in a scalable way
4. The building of models for event recognition using partial data from each sensor, in particular to deal with the spatial and temporal distribution of the incoming data.
5. Building scalable models for event recognition, in particular with ability to detect abnormal events
6. The building of models with limited data, in an incremental fashion.
7. The learning of event patterns in terms of multi-modal distributed sensor data with particular emphasis on building closed loop, self configurable systems
8. The construction of tools and technologies for incrementally updating notions of abnormal and normal
9. Exploration of the extent to which the user can be used to guide the system in building models with limited data
10. The exploration of use of multiple modalities in event recognition, in particular with an emphasis on developing methods to use modalities incrementally to develop an escalating warning system
11. The need to build a self configurable, calibrating sensor group, in the case when the sensor location is fixed and when the sensors are mobile.

These techniques can be used in multiple domains, and in particular we explore the connection to the UAV project. The UAVs can be equipped with multiple sensors; cameras, infrared cameras, etc. Each sensor will have limited processing power, and the key challenge will be to develop a pattern recognition system in this domain. The sensors will be geographically separated and possibly multi-modal. The challenge to recognize abnormal and normal events then translates into solving the problems identified above.