I am taking final year's projects supervision this year. If you have an idea that fits into robotic vision, computer vision, machine learning, or all of them, drop me a line so we can arrange to meet and discuss it.

New projects this year (2018-2019)!

Here are some new project ideas I would like to run this year:

I. Puzzles

Humans are good at solving puzzles, and children as young as 2-years-old can solve simple ones. There are a number of Psychological theories on how we may be doing this (using pictorial vs shape cues). This project would have two aims: 1) devise a computer vision approach for automatically solving puzzles; and 2) investigate what types of features and representation of visual information are efficient for puzzle-solving.

II. What makes a picture

The aim of this project would be to use a machine learning and computer vision approach to analyse artistic graphic styles from a data driven perspectives. The aim would be to discover characteristics of artworks and be able to predict them on new pictures.

III. Robot pundit: Automatic analysis of sport footage

This project will aim at analysing broadcast footage of sport events, to train a system to detect key events and ultimately provide a live commentary. The project will focus on the game of rugby, which contains an interesting combination of local individual actions (eg, kicking the ball) and more global, team-wide events (eg, scrums). This project involves a significant programming component, C/C++ knowledge, and from the theoretical standpoint will touch aspects of computer vision and machine learning.

IV. This place looks familiar.

This project will aim to develop a "visual GPS": estimating the location from a single camera image. The project will make use of Google Streetview API to collect images of Exeter, and attempt to build a model to provide geographic location from visual features only. Again, this project involves a significant programming component, C/C++ knowledge, and from the theoretical standpoint will touch aspects of computer vision and machine learning. See this paper for a similar project.

Previous years' projects (but still fresh ideas!)

Additionally, Here are some project ideas from last year (still up for grab - come and see me to discuss what was achieved last year!):

I. Fast & Cautious: Learning autonomous driving on a simulator

Last year a very successful project [1] devised a program that learnt to steer around a racing track in a simulator using visual information only, while keeping the car's speed constant. This year's project will look at how a system can learn to adapt its speed in order to allow for the fastest lap times - while still managing to steer around the curves. This project involves a significant programming component, C/C++ knowledge, and from the theoretical standpoint will touch aspects of computer vision and machine learning.

[1] Reinis Rudzits (2014). Learning Autonomous Driving in a Racing Simulator. BEng Electronic Engineering Thesis. (pdf)

II. Are you looking at me?

Humans are extremely good at estimating what another person is looking at. The aim of this project is to devise a system that can estimate the direction of attention of characters in videos. The project will entail gathering and annotating a few sequences, eg, from soaps or talk shows, detecting faces using existing algorithms, and attempting to learn a predictor for the character's focus of attention. This project involves a significant programming component, C/C++ knowledge, and from the theoretical standpoint will touch aspects of computer vision and machine learning.

III. How to play Mario?

Recently published results have demonstrated that it is possible for a computer to learn autonomously to play video games, although the proposed approach that learns playing from scratch using trial and error requires significant computational power. This project proposes to investigate how a computer can learn basic playing skill by imitating a human player, the aim will be to learn from the human player significant patterns (ie, dangers), versus irrelevant ones (ie, background art). As ever, this project involves a significant programming component, C/C++ knowledge, and from the theoretical standpoint will touch aspects of computer vision and machine learning.

IV. Body pose estimation from videos

The aim of this project is fairly simple in appearance: to estimate the body pose of subjects in videos. Unfortunately, as often with computer vision it turns out to be a difficult problem to solve automatically. There has been many good projects attempting to solve this difficult problem, but this remains an unsolved interesting problem. The project will mainly involve machine learning work to learn what visual patterns can predict the locations of body parts.