Safety in Smart Vehicle - Pedestrian Interaction

There has been a lot of research on smart vehicles (SV, including autonomous vehicles and smart powered wheelchairs), mainly for motorways and other structured environments, with resulting safety levels in such highly structured conditions being excellent.

However, the situation is different for less structured environments, particularly where interaction between SV and pedestrians is possible, such as pedestrian junctions and mixed traffic environments.

In these cases, more fundamental research in safety aspects is needed, since even minor contact between humans and vehicle poses serious dangers to unprotected humans. Specifically, pedestrian intention prediction is crucial for safe and smooth SV operation.

This project aims to develop multi-source and multi-modal algorithms which can predict intentions of pedestrians under challenging lighting conditions (using both visible (RGB) and thermal imaging), using cues from both pedestrian movements as well as their environmental and social context. The project aims at enhancing the safety level of pedestrians in the context of SV in unstructured environments.

Apart from the development of novel algorithms in this challenging domain, we aim to maximise the impact of our research through the creation of one of the first pedestrian intention prediction datasets combining RGB and thermal images.

Performance evaluation of intention pedestrian algorithms will involve both vehicles in intersections, as well as smart wheelchairs for people with disabilities.

By enhancing the safety level of SV and pedestrians through predicting pedestrians' intention under various lighting conditions, the results of this project will be very helpful for the development of SV, and will also promote the public' s acceptance of SV.

Consequently, the results of this project are very beneficial for the EU, where multiple governmental and commercial autonomous driving initiatives are active.