Nigel Whyte

BSc MSc MIEEE

Lecturer II Computing Department
 
  • Research Interests
  • Publications
  • Research Supervision

Research Experience

A Hand Gesture Recognition System

The ability to recognise particular gestures is essential if computers are to interact naturally with human users. At present this information is unavailable to machines. This relatively new application of computer vision aims to extend the interface between man and machine.
 
We are interested in developing algorithms without the use of external devices for sign language recognition. This recognition consists of difficult computer vision problems: Hand Identification and Gesture Recognition/Sign Identification.
 
Hand Identification: Initially the hand must be identified or segmented from the rest of the image. Here we will investigate current algorithms used for object detection and employ colour information in our final algorithm.
 
Gesture Recognition/Sign Identification: Here we need to identify the hand gesture/sign language. Template matching will be used to identify the gesture/sign. This is a non-trivial problem, as the finger position will be different for the same ‘language signs’. To this end neural networks and depth information will be used to solve this problem.
 
Finally an efficient ‘Sign Language’ identification algorithm for visual images will be delivered. Such an algorithm would be useful in other applications including human computer interface, machine vision and machine control in industry.
 
Person Identification Using Image and Voice Recognition
We identify people based on their facial characteristics, among others. The idea here is that a person has their image taken by a digital camera, the image is segmented and their characteristics extracted (using computer vision and neural network techniques). These characteristics are compared to a database of people to identify the person. However, using just facial features alone will not give 100% success rate (people change the colour of their hair, grow/remove facial hair etc.).  To overcome this we include voice recognition as well to improve the accuracy.
 
Three-dimensional Reconstruction From Stereoscopic Images
The measurement of depth is an essential skill for interacting with our environment. All humans have two eyes that are laterally separated by about 65mm. Even with this slight separation, we get a different view of the world from each of our eyes. The different perspectives of our eyes creates slight displacements of objects called disparities, in the two monocular views of the scene. The human visual system is able to use these disparities for depth-estimation and merge both monocular views to get a 3D view of the scene. The correct (solving the correspondence problem) and efficient computation of disparities is a difficult problem in computer vision.
 
Problems occur due to the real world environment such as light for example, specula reflections that move independently of the surfaces of objects. Common computational approaches include feature-based, area-based, and phase based methods. 
All these methods have their problems, here these techniques are being implemented and explored to find a new dimension to the successful solving of the problem. 
 
Investigating Dual Delivery – Combined CD-ROM and Internet Delivery
Multimedia applications running on networks were/are drastically affected by the delivery rate problems caused by the lack of bandwidth. Dual delivery is a system where an application is run on the computer using a CD and the internet for content delivery. It proved an excellent delivery method for multimedia as it combined the best of both CD and the internet. The CD is used to store the large files that are time sensitive (e.g. video and sound) while the network is used to deliver smaller files requiring low bandwidth.
 
In this research software tools are developed that allowed a developer to create dual delivery multimedia applications without the need to reference the delivery medium protocols.

 

Using Colour Information to Solve the Correspondence Problem 
In the late 1980’s computer vision researchers had only begun investigating the use of colour information in their study. Due to greatly reduced hardware costs, colour image processing had become an integral part of many scientific and industrial applications. 
 
Colour greatly affects our ability to differentiate among objects. In this research, colour information is used as a clue to solving the correspondence problem in Stereopsis. Evidence is provided of the use of colour information in the stereo correspondence process in the human visual system upon which the final algorithm is based. The colour dependent system consists of edge detection, feature detection and matching, and the solving of the correspondence problem.
Knot Identification in Tree Cross Sections
 
The quality of timber is determined on several factors from plantation site to the physical structure of the tree. Here we are concerned with the physical structure of the tree, which is dependent on defects within the log, particularly knots that weaken the timber and consequently reduce its quality. Recently, computer tomography was investigated to facilitate the identification of defects. In this way information on the internal structure and characteristics of the log is gathered. Here we are concerned with the identification of knots, as they are the most common internal characteristic that reduces the quality of planks. We are investigating the use of various computer vision and neural network techniques to effectively identify the knots.
 
Object Tracking
The development of algorithms for tracking objects such as cars is an active area of research with potentially several uses in our every day life. It is a complex task given that the following may occur over the series of images: background changes, the profile of the object, possibly of occlusion, and intensity changes. 
 
Traditionally features have been used to track objects over a sequence of images. Such systems have proved problematic due to false object detection and the processing time required to evaluate an image frame. In this research, we are investigating traditional algorithms with the use of colour information and neural networks to provide a more accurate and efficient object detection process.
 

Funded Projects & Research Interests

Funded Projects

  • Department of Education and Science Post-Graduate R&D Skills Programme, A Hand Gesture Recognition System, 2001, £22,500. 
  • Enterprise Ireland Applied Research Grant, Three-dimensional Reconstruction from Stereoscopic Images, 2000, £34,200. 
  • Enterprise Ireland Applied Research Programme, An Investigation into the use of Colour Information for Object Tracking, 1998, £5,000. 
  • Multimedia Technology Ireland, Distributed Multimedia - Combined CD-ROM Internet Delivery, 1995, £10,000. 

Research Interests

Nigel's research interests are mainly in computer vision including stereopsis, object recognition and tracking, pre-processing segmentation, and neural networks. He has worked as a consultant for several multinational companies on computer vision related projects.
 

Publications

  • G. Somers, R.N. Whyte. Hand Posture Matching for Irish Sign Language Interpretation. In Press. 
  • R.N. Whyte, G. Somers. ISLIS: Irish Sign Language Identification System, Irish Machine Vision and Image Processing Conference, University of Ulster, Coleraine, Ireland, pp.33-40, 2003. 
  • Y. Vanderstockt, R.N. Whyte. Watershed Transformation: Reducing The Over-Segmentation Problem by Applying a Noise Reducer and a Region Merger, In Proceedings of The 10th International Conference in Central Europe on Computer Graphics, Visualization, and Computer Vision, Plzen, Czech Republic, Vol. 4, pp. 57-60, 2002. 
  • G. Somers, R.N. Whyte. Identifying Hand Movement Patterns in Three