Intelligent Systems Group - Research Focus

Developing systems that can operate in dynamic and unpredictable environments requires advances beyond that those typically possible in isolated knowledge fields. Systems found in nature appear to seamlessly integrate concepts from a wide range of knowledge fields that has proven challenging to mimic and has been the source of inspiration for many innovations. In addition, animals and humans incorporate various levels of reasoning and can interact with their environment. The aim of the Intelligent Systems Group (ISG) is to develop complex systems that not only incorporate various knowledge fields, but exhibit the ability to reason about, and interact with, the environments they must operate in. Although developing truly intelligent systems is the overall goal of the ISG, the secondary goal is to incrementally apply advances in the theory of intelligent systems to existing systems. This secondary goal often reduces to incorporating algorithms from the fields of Artificial Intelligence and Machine Learning to classical systems approaches.

The research focus of the ISG therefore lies in the theory and methods whereby various degrees of intelligence can be simulated in systems. Although intelligence can be applied in various real-world applications, and the ISG activities include such applications, the focus of the ISG is the development of new approaches to simulated intelligence. The assumption is that such new approaches will be tested in various applications, ranging from the obvious such as robots, to the less obvious such music and planning systems. The various applications and specific systems being developed therefore represent particular test cases of more general theory and methods being developed.

The primary focus areas of the group currently are listed below. Prospective students interested in the general topic areas should contact the corresponding researcher for details of the research, possible research topics and to discuss opportunities (e.g., funding).

Data fusion

  • Overview: The research group within the data fusion focus area investigates theoretical aspects and applications in the field of data fusion, specifically those aspects that have importance in a South African context. The group also aims to address a lack of expertise in the field of data fusion in South Africa. Data fusion involves combining information and the extraction of patterns — in the broadest sense — to estimate or predict the state of some aspect of the universe, such that the resulting combined information is richer and more informative for the objective at hand. It encompasses not only the combination of lower level information such as multiple sensor data, but also high level (processed) information. The theoretical basis for such methods is the Bayesian statistical approach, which involves the combination of prior knowledge with sensed data. The aim of the research being performed by the ISG is to develop methods to correctly integrate the various data streams available and to detect and extract patterns. Part of the formal definition of data fusion includes sensor resource management which results in active changes to the sensing process. The detection and extraction of patterns, in turn, typically requires techniques from the field of pattern recognition. 

Digital Image Processing and Computer Vision

  • Overview: Digital Image Processing (DIP) entails the manipulation of images in the form of 2-dimensional matrices of pixel values. The most common operations are that of image enhancement and noise reduction. The images under consideration are not limited to the visual spectrum but may consist of multiple spectral bands. Whereas digital image processing focuses on the low level operations on images, computer vision focuses on the extraction and interpretation of information from images. The most common operations entail the extraction of 3D information from one or more 2D images and the recognition and interpretation of the objects contained in images. As vision is one of the most powerful senses of humans, the combination of image processing and computer vision represents one of the most powerful sensing approaches available to artificial systems. However, there is currently only limited understanding of the human vision system and correspondingly there is still significant scope for the advancement of the theory and methods of image processing and computer vision. The primary area currently being investigated by the ISG is real-time 3D ridged and non-ridged object recognition and pose estimation as part of robotic systems.
  • Primary contact: Mr H Grobler <hans.grobler at up.ac.za>

Music and Artificial Intelligence

  • Overview: There is at present strong interest in developing systems which can recognise music, interpret/transcribe/analyse music, perform music and also create music. Recognition of music has become especially relevant in view of the large quantities of music that is currently available in digital format, for example for download from the Internet. Pertinent topics range from low-level feature extraction, such as real-time pitch tracking from the raw audio signal, to the use of machine learning and pattern recognition techniques to recognise genre, mood, and performer — a practical application being intelligent playlist compilation or recommendation systems. Intelligent systems that involve music making at some level have included interactive accompanying systems, as well as automated composition algorithms accommodating various styles. The latter typically incorporates approaches based on stochastic process models, grammars, genetic algorithms, neural networks, etc. The current main focus within the ISG is on music recognition/interpretation systems, in the recognition case focusing on genre and performer. This allows for specific research topics that might be defined in accordance with students' musical interests and theoretical background. 

 

Published by Hans Grobler

Copyright © University of Pretoria 2021. All rights reserved.

FAQ's Email Us Virtual Campus Share Cookie Preferences