Parametric tracking with spatial extraction across an array of cameras

Video surveillance is a rapidly growing area that has been fuelled by an increase in the concerns of security and safety in both public and private areas. With heighten security concerns, the utilization of video surveillance systems spread over a large area is becoming the norm. Surveillance of a large area requires a number of cameras to be deployed, which presents problems for human operators. In the surveillance of a large area, the need to monitor numerous screens makes an operator less effective in monitoring, observing or tracking groups or targets of interest. In such situations, the application of computer systems can prove highly effective in assisting human operators.
The overall aim of this thesis was to investigate different methods for tracking a target across an array of cameras. This required a set of parameters to be identified that could be passed between cameras as the target moved in and out of the fields of view. Initial investigations focussed on identifying the most effective colour space to use. A normalized cross correlation method was used initially with a reference image to track the target of interest. A second method investigated the use of histogram similarity in tracking targets. In this instance a reference target’s histogram or pixel distribution was used as a means for tracking. Finally a method was investigated that used the relationship between colour regions that make up a whole target. An experimental method was developed that used the information between colour regions such as the vector and colour difference as a means for
tracking a target. This method was tested on a single camera configuration and multiple camera configuration and shown to be effective.
In addition to the experimental tracking method investigated, additional data can be extracted to estimate a spatial map of a target as the target of interest is tracked across an array of cameras.
For each method investigated the experimental results are presented in this thesis and it has been demonstrated that minimal data exchange can be used in order to track a target across an array of cameras. In addition to tracking a target, the spatial position of the target of interest could be estimated as it moves across the array.