My research within Cyberscout and in general involves pattern recognition, machine learning and multi-agent systems. I am currently developing an autonomous visual surveillance system executing on mobile and stationary platforms. The surveillance system uses a multi-agent architecture designed to elicit emergent behaviors. The focus thus far has been in developing low level algorithms to aid in visual surveillance. I have developed an efficient motion detection algorithm that can segment moving objects from streaming video in real-time while the camera is stationary. I am now developing an artificial neural network based correspondence algorithm that can temporally associate targets within a single camera and also within a network of cameras on multiple platforms. Each algorithm is executed by an agent. Collaboration between agents (algorithms) such as the detection and correspondence algorithms increases the performance of the system.
I am also interested in abnormal event detection for visual surveillance. One aspect of this interest (with many other applications) is automatic discriminative feature detection. I currently use this to efficiently track a moving object within a sensor and across sensors (this is also sometimes called 'target hand-off'). Automatic discriminative feature identification also has applications in content based retrieval and fault detection and diagnosis.
As part of my work in target correspondence, feature identification and content based image retrieval, I have developed a technique called Differential Discriminative Diagnosis. This technique is described in detail in my Master's Thesis.
|Research Interest Keywords|
|computational neuroscience, machine learning, pattern recognition|
|The Robotics Institute is part of the School of Computer Science, Carnegie Mellon University.|
Contact Us | Update Instructions