• home
• current students
• graduate
• events

Presenter:
Supervisor:

Date: Tue, March 8, 2005
Time: 14:30:00 - 15:30:00
Place: EOW 430

ABSTRACT

Abstract

Computer vision has started as a subfield of Artificial Intelligence. The original purpose of computer vision was the development of algorithms for automatic image understanding. Specifically, these algorithms were designed for the detection, segmentation, description and recognition of certain objects in static images. Such precise tasks may find a straightforward implementation in robotic systems, designed to interact with simple, controlled environments. However, in more complex situations such as medical applications, the design of systems for computer-aided diagnosis has to search for the optimal trade-off between the amount of human and artificial intelligence involved in a decision making process. Therefore, the focus shifts from automatic image understanding towards finding appropriate paradigms for user interaction and knowledge modelling.

During the first part of the seminar, we will present our recent research work related to medical applications. Among other examples, we will discuss our contribution to the design of a virtual environment for the simulation of liver cancer cryotherapy (see Figure 1 for more details).

Figure 1. Two different views in the visualization of the liver, the tumour (in blue) and the major liver vessels (in orange).

The second part of our presentation will briefly describe our recent work in computer-vision based human motion analysis. Understanding human motion is a non trivial task, useful for applications such as the design of multimodal user interfaces based on natural travelling metaphors. Intelligent systems for surveillance may also embed techniques for pedestrian tracking and for gait-based person identification. In this context, we will discuss our contribution to the development of a computer vision-based monitoring system for nonintrusive and real-time tracking of persons moving around in public or private premises (see Figure 2 for more details).

Figure 2. Feature-based tracking from a monocular video sequence.