The main aim of this PhD is to investigate how a methodology rooted in systems engineering concepts can be established and applied to the design of distributed wide-area visual surveillance systems. Nowadays, the research community in surveillance systems tends to be mostly focused on the computer vision part of these systems, researching and developing more intelligent algorithms. The integration and finally the creation of the system per se, are usually regarded as a secondary priority. We postulate here that until a robust systems-centred, rather than algorithmic-centred approach is used, the realisation of realistic distributed surveillance systems is unlikely to happen. The future generation of surveillance systems can be categorised, from a system engineering point of view, as concurrent, distributed, embedded, real time systems. An important aspect of these systems is the inherent temporal diversity (heterogeneous timing) that arises from a variety of timing requirements and from the parallelisation and distribution of the processes that compose the system. Embedded, real-time systems are often naturally asynchronous. However, the computer vision part of these surveillance systems is commonly conceived and designed in a sequential and synchronous manner, in many cases using an object-oriented approach. Moreover, to cope with the distributed nature of these systems, technologies such as CORBA are applied. Designing processes in a synchronous manner plus the run-time overheads associated with object oriented implementations may cause communication bottlenecks. Perhaps more importantly, it may produce unpredictable behaviour of some components of the system and hence undetermined performance from a system as a whole. Clearly, this is a major problem on surveillance systems that can often be expected to be safety-critical. This research has explored the use of an alternative approach to object-orientation for the design and implementation of intelligent distributed surveillance systems. The approach is known as Real-Time Networks (exemplified by system engineering methodologies such as MASCOT and extensions such as DORIS). This approach is based conceptually on conceiving solutions as being naturally concurrent, from the highest level of abstraction, with concurrent activities communicating through well-defined data-centred mechanisms. The methodology favours a disciplined approach to design, which yields a modular structure that has close correspondence between functional elements in design and constructional elements for system integration. It is such characteristics that we believe will become essential in overcoming the complexities of going from small-scale computer vision prototypes to large-scale working systems. To justify the selection of this methodology, an overview of different software approach methods that may be used for designing wide-area intelligent surveillance systems is given. This is then, narrowed down to a comparison between Real-Time Networks and Object Orientation. The comparison is followed by an illustration of two different design solutions of an existing real-time distributed surveillance system called ADVISOR. One of the design solutions, based on Object Oriented concepts, uses CORBA as a means for the integration and distribution characteristics of the system. The other design solution, based on Real-Time Networks, uses DORIS methodology as a solution for the design of the system. Once the justification over the selection is done, a' novel design of a generic visual surveillance system using the proposed Real-Time Networks method is presented. Finally, the conclusions and future work are explained in the last chapter.

Palmar flexion crease matching is a method for verifying or establishing identity. New methods of palmprint identification, that complement existing identification strategies, or reduce analysis and comparison times, will benefit palmprint identification communities worldwide. To this end, this thesis describes new methods of manual and automated palmar flexion crease identification, that can be used to identify palmar flexion creases in online palmprint images. In the first instance, a manual palmar flexion crease identification and matching method is described, which was used to compare palmar flexion creases from 100 palms, each modified 10 times to mimic some of the types of alterations that can be found in crime scene palmar marks. From these comparisons, using manual palmar flexion crease identification, results showed that when labelled within 10 pixels, or 3.5 mm, of the palmar flexion crease, a palmprint image can be identified with a 99.2% genuine acceptance rate and a 0% false acceptance rate. Furthermore, in the second instance, a new method of automated palmar flexion crease recognition, that can be used to identify palmar flexion creases in online palmprint images, is described. A modified internal image seams algorithm was used to extract the flexion creases, and a matching algorithm, based on kd-tree nearest neighbour searching, was used to calculate the similarity between them. Results showed that in 1000 palmprint images from 100 palms, when compared to manually identified palmar flexion creases, a 100% genuine acceptance rate was achieved with a 0.0045% false acceptance rate. Finally, to determine if automated palmar flexion crease recognition can be used as an effective method of palmprint identification, palmar flexion creases from two online palmprint image data sets, containing images from 100 palms and 386 palms respectively, were automatically extracted and compared. In the first data set, that is, for images from 100 palms, an equal error rate of 0.3% was achieved. In the second data set, that is, for images from 386 palms, an equal error rate of 0.415% was achieved.

This thesis describes the development of methods for automatically obtaining coarse gaze direction estimates for pedestrians in surveillance video. Gaze direction estimates are beneficial in the context of surveillance as an indicator of an individual's intentions and their interest in their surroundings and other people. The overall task is broken down into two problems. The first is that of tracking large numbers of pedestrians in low resolution video, which is required to identify the head regions within video frames. The second problem is to process the extracted head regions and estimate the direction in which the person is facing as a coarse estimate of their gaze direction. The first approach for head tracking combines image measurements from HOG head detections and KLT corner tracking using a Kalman filter, and can track the heads of many pedestrians simultaneously to output head regions with pixel-level accuracy. The second approach uses Markov-Chain Monte-Carlo Data Association (MCMCDA) within a temporal sliding window to provide similarly accurate head regions, but with improved speed and robustness. The improved system accurately tracks the heads of twenty pedestrians in 1920x1080 video in real-time and can track through total occlusions for short time periods. The approaches for gaze direction estimation all make use of randomised decision tree classifiers. The first develops classifiers for low resolution head images that are invariant to hair and skin colours using branch decisions based on abstract labels rather than direct image measurements. The second approach addresses higher resolution images using HOG descriptors and novel Colour Triplet Comparison (CTC) based branches. The final approach infers custom appearance models for individual scenes using weakly supervised learning over large datasets of approximately 500,000 images. A Conditional Random Field (CRF) models interactions between appearance information and walking directions to estimate gaze directions for head image sequences.