Pieter de Villiers, a professor in the Department of Electrical, Electronic and Computer Engineering in the University of Pretoria’s Faculty of Engineering, Built Environment and Information Technology, delivered his inaugural address on 4 October 2022. The theme of his address was “Seeing a dynamic world through the eyes of sensors”.

Prof De Villiers heads the Department’s Signal Processing and Telecommunications Research Group, and is Co-Chair of the MultiChoice Chair of Machine Learning. The focus of his research includes statistical signal processing, machine learning (with a focus on signal processing), sensor and data fusion, and Bayesian inference.

The application areas of his current research include audio and video processing, radar, financial management and machine condition monitoring. It has an impact across several of the Faculty’s research focus areas, as well as two of the United Nations’ Sustainable Development Goals (SDGs): SDG 9: Industry, innovation and infrastructure, and SDG 15: Life on land.

According to Prof De Villiers, the fusion and interpretation of data from multiple sources and sensors are important for automating complex tasks, such as tracking pedestrians or aircraft, or coordinating robots or autonomous vehicles in complex and dynamic environments. During his public lecture, he introduced a philosophical view of the sensing problem, how it is formalised in Bayesian probability, how the Bayesian models are used to make inferences and predictions, and how these inferences are then used to take action.

Prof De Villiers explained that two things are needed to make inferences: models and data. “Current research in the field focuses on the parallelisation of information fusion algorithms, and uncertainty representation and reasoning in information fusion,” he said. “Other efforts focus on the intersection of machine learning and information fusion, as well as concepts of explainability and trust in these methods.”

Due to the transdisciplinary nature of the problems that need to be solved by adopting Bayesian and other probability methods, the solutions developed through signal processing and machine learning have many application areas.

Prof De Villiers discussed a few of the applications and use cases addressed in his research. Safety and security use cases include the classification of humans and animals using Doppler radar for counter-rhino poaching operations and behavioural modelling for the classification of maritime vessels for anti-abalone poaching and anti-piracy. Recently, he has been considering using Bayesian neural networks to quantify uncertainty in synthetic aperture radar image for target detection, as well as to improve training in scarce data applications.

He detailed further use cases of his research efforts in broader application domains, such as video broadcasting and video streaming, financial risk modelling, machine diagnosis and prognosis, and the modelling of gene expression lineages. Although this field is gaining increasing traction, researchers are experiencing some challenges. These include the computational efficiency brought about by parallelisation, where algorithms are split between processors, and the use of machine learning for modelling, where the target behaviour needs to be learnt. Furthermore, when human lives are at stake, well placed trust in the answers provided by machines is essential, and needs to be in accordance with reliability and performance.

Prof De Villiers concluded his presentation with a future vision for sensor and data fusion research, where intelligent sensing in machine automation is becoming increasingly prevalent.