Abstract:

The applications of multi-target detection and tracking are of crucial importance in defence applications and robotics. Recent advances in multiple-target filtering provide a general systematic treatment of multi-object filtering using random sets. These have led to robust multiple-target tracking algorithms, known as Probability Hypothesis Density (PHD) filters, for estimating both the correct number of targets and their positions in complex environments with large numbers of false detections and where the true targets are not always detected. These can be exploited on a range of sensor systems without complex target detection algorithms. Some practical applications of these techniques have been developed for tracking in sonar for autonomous vehicle navigation, concealed weapons detection in millimetre wave imaging, people tracking in video, and audio signal processing for music transcription. This talk will present recent developments in this field and describe some of the current research challenges being investigated.

About the Speaker:

Daniel Clark was awarded his PhD on Multiple Target Tracking at Heriot-Watt University , Scotland in October 2006. He worked as a Research Associate in Cambridge before returning to Heriot-Watt as a Lecturer in September 2007.

Part 2: Articulated Body Pose Estimation and Modelling in Disparity Space

Immersive videoconferencing aims to achieve a sense of tele-presence during a videoconferencing session in which the participants are geographically distributed around the world. The limitations of the setup require the wide-baseline stereo pairs of cameras to be placed around the screen, providing a viewpoint which is different from the one expected by the conversation partners.

Image-based novel-view synthesis is a technique used to render the remote conference participants from the correct viewpoint and in this way strengthen the sense of presence. The quality of the view synthesis depends on the quality of the correspondence information (disparity maps) available in the stereo setup.

 We investigated the use of a generic human body model to improve the quality of the disparity information for novel view synthesis. In this context, we addressed the problem of articulated human body pose estimation with particle swarm optimisation and articulated human bod modelling in disparity space for the purpose of disparity data completion. In the talk we will present recent results of this research.

About the Speaker: