All the projects are primarily software or involve systems type integration with existing pieces of hardware (such as the RT100 robot arm, cameras and frame-grabbers, robot vehicles etc). The language for programming is either C or C++. Previous knowledge of this is not essential: it can be learnt as part of the project. There are no pre-requisites modules needed for any of the projects although some relevant material is likely to be covered in the Machine Intelligence final year module.
This project concerns the ability of a mobile agent to simultaneously find where it is in an environment whilst also building up a model of that environment. This involves finding and tracking images feature in 2D and then inferring their 3D location. This is a challenging task for many reasons (data errors, changes in appearance of features with change in viewpoint etc) but if it can done in a tight control loop then it would enable a robot to formulate an awareness of its environment and enable it to begin planned navigation of such an environment. This is a challenging project with much scope and would be well suited to a good student. then
This is a challenging project which aims to build a hand held laser scanning device that can be used to construct graphical models of 3D objects. The basic principle of laser scanning is well established but such scanners are usually precision instruments in which the sensor location is accurately determined by mechanical means. The challenge of this project is to simultaneously estimate both the shape of the external object while at the same time measuring by visual means the position of the sensor. A recent article by a Canadian group has deminstrated the basic feasibility of this and would form a starting point for the project. The project is quite difficult and also involves a lot of work and therefore requires a first class student.
Wearable electronics is currently a topic of great interest. It is possible to incorporate small cameras into clothing and acquire real-time imagery of the world. This project will consider how specific objects can be recognised from such video data. Such objects could be signs on buildings which would allow a person to identify their location. A computer system with this capability could form the core for an automated tourist guide system that would recognise what the wearer was viewing and then provide the user with factual information about interesting sights. The project will focus on recognition of objects from contour information using techniques that are robust to affine effects that result from viewpoint changes.
A camera provides a spatially limited view of the world. However, if a camera is mobile, either by being on a moveable head or on a mobile platform, then it should be possible to combine the information from several images to construct a larger view of the world. Such capabilities is useful in many areas. This project will look at existing work on creation of large video mosaics and will implement and test one of the promising techiques.
An image is a 2D view of a 3D world. Teleoperation of a mobile vehicle involves using camera data to navigate across potentially difficult outdoor terrain. In addition to live camera data, a tele-operator may have other information about the region that is navigated. For example, the Ordanace Survey provide Digital Elevation Maps for most of the UK. This project considers how live image data and Ordanance Survey data can be used in co-operation to produce a better system for driving a tele-operated vehicle.
Not all information in a an image is equally important. The human vision system extracts little information from individual images but manages to build a consistent and persistent model of the world. This project will explore ideas about how this is achieved and look to implement some algorithms that address this difficult but fundamenatal problem. The project will integrate work from Psychology, NeoroScience, Computer Science and Artifical Intelligence.
A single static camera provides limited information about the world. However, the addition of a set of mirrors (which may be able to rotate) allows a single camera to gather a set of views of a scene. This set of views can be used for tasks such as stereo vision. This project will involve a mix of hardware and software to construct a mirror based imaging system suitable for mounting on a modile vehicle. The system will be used to construct a 3D model of the world.
Professor John Illingworth