Environmentally robust multiple camera tracking

A significant growth of the use of surveillance cameras has arisen from both the availability of low-cost home security and post-September 11th security measures. With such a plethora of surveillance cameras available and already in use, tracking a person or object from one field of view to another accurately is a challenging possibility; recognising the same person at different spatial locations, under different lighting conditions, at different scales and orientations. In order to address these challenges and provide a solution, a review of recent and past literature is provided. The main theme of this research is investigating methods to improve tracking of objects and people in dynamic environments and applying computational techniques to provide solutions to optimise such tracking systems. Image processing techniques are explored and refactored to adapt to currently available single-board computing power. Optimisation methods for speed of computing are investigated, presenting the paradigm of parallel programming during the design of “computationally intense” algorithms. The research also addresses cross-platform software/ server application design. In controlled environments current tracking systems perform well, however, this project explores methods to take multiple camera tracking to a higher level where they can, in real time, robustly cope with: rapid changes in lighting and track objects between indoor and outdoor scenarios at any time of day or in any weather conditions, severe image occlusion, rapid changes in direction, orientation and velocity of the object being tracked and be invariant to image clutter and noise. Thus the outputs are twofold: track a human/object across multiple cameras and ensure the algorithm is fast enough to run in real time on a modern processor. This research explores algorithms to deliver colour illumination invariance, also known as colour constancy. Colour illumination invariance can be applied as a pre-processing step to all cameras in a multi-camera environment. The research also investigates experimental assessment of multi-camera performance, focusing mainly on robustness to environmental changes. There are three main objectives for a tracking algorithm being used in the proposed system. Firstly, the tracking algorithm must accurately detect objects independently of their scale change and rotation. Secondly, the tracking algorithm must accurately detect objects across multiple cameras in different lighting conditions. The third objective for the tracking algorithm is that it must be able to attain a high level of colour constancy. The last objective can be implemented as a pre-processing step to such a tracking algorithm. This research explores the use of the Scale Invariant Feature Transform (SIFT) and the Speeded-Up Robust Features (SURF) algorithm. These algorithms are discussed in detail in the literature review as well as methods for providing colour illumination invariance.