We consider the dense mapping problem where a mobile robot must combine range measurements into a consistent world-centric map. If the range sensor is mounted on the robot, as is usually the case, some form of SLAM must be implemented in order to estimate the robot's pose (position and orientation) at every time step. Such estimates are typically… (More)
We investigate the feasibility of using the Radon transform and a dynamic programming algorithm to authenticate handwritten signatures on checks. Since there no dynamic information is available as in the case of the online problem where signatures are typically captured by means of a digitising tablet, the ooine problem poses serious challenges. Our present… (More)
The aim of this project is to develop a robust and effective face (head) tracker based on the Kanade-Lucas-Tomasi (KLT) tracking algorithm. The project concentrates mainly on robust tracking despite excessive background clutter.The problem of disappearing features and the re-generation of features on a persons head is also investigated. The system can also… (More)
Date The final copy of this thesis has been examined by the signatories, and we find that both the content and the form meet acceptable presentation standards of scholarly work in the above mentioned discipline. The radial basis functions (RBF) method is meshfree, easy to implement in any number of dimensions and spectrally accurate for certain types of… (More)
We propose a multi-estimation algorithm based on the Unscented Kalman filter, that estimates the structure and motion of an object from a video sequnece under perspective projection. It is shown that the algorithm is stable and accurate, requiring no prior initialisation.
Recognizing text in natural images can be a useful tool for image understanding. We focus on the detection problem, which is to find regions in an image occupied by text. We consider multi-layered convolutional neural networks as a means to classify local regions as text or not, and take a sliding-window approach to scan a full image. For training we… (More)