Underwater Imaging and/or Predicting Behaviour on Urban Roads

As part of an EPSRC Funded Programme, we are working with the Universities of Birmingham and Edinburgh and Jaguar Land Rover to explore new technologies to pave the pathway for a fully or partially autonomous driving, developing sensing systems are required to enable automatic optimisation of the vehicle for travel over any terrain in all weathers.
(https://www.epsrc.ac.uk/newsevents/news/jlrannouncesautonomousvehicalresearchprogramme/ )
We are now looking for a talented individual to join the team for a fully funded PhD programme. The candidate will have a 1st class undergraduate degree (or equivalent) in a mathematics, physics or engineering course and be strongly motivated to pursue research as part of an existing team. He/she should have excellent mathematical abilities, and experience in aspects of signal and image analysis. Excellent skills and experience of computer programming in C/C++ and/or MatLab would also be highly advantageous. However, for an excellent candidate, there are opportunities to enhance these skills early in the programme.
There are two possible areas of study. First, we wish to investigate how to combat the dangers of driving through flood water, standing water or fords when hidden objects or depressions may be present, and tyres may lose contact with the road and the driver loses steering control. We would investigate full waveform LIDAR technology and sonar to perform urban flood underwater imaging in a window of approximately 1-10m in front if the vehicle to a depth of approximately 1m. The system should detect and classify impediments to progress, for example potholes, objects lying on the ground such as fallen branches and rocks, and potentially even floating objects.

Second, we could examine how to predict actor (other cars, pedestrians etc.) behaviour when detected and tracked from a mobile vehicle. For example, how do we tell if a pedestrian is about to step off the pavement in front of a vehicle, or a vehicle is about to change lanes in a dangerous manner. To that end, we wish both learn and predict behaviour from deep analysis of what is sometimes called ‘pattern of life’. However, the key difference from earlier work is that the data is dynamic and uncertain, and one has to take into account the results of on-going data processing, including problems of mis-detection and mis-classification in deciding what to do next.