Multi camera surveillance systems is a hot topic in computer science, artificial intelligence and security management in general. While cameras used to be low resolution analogue devices, nowadays they stream high definition images and come equipped with the processing power to reason about their environment. The field covers a wide variaty of researches from streaming large amount of data to reasoning about the behaviour. In this thesis we will address the problem of cooperation and task delegation within a distributed 3rd generation visual surveillance system. Multiple intelligent cameras will work together to solve the problem of detecting, tracking and analysing the behaviour of cars in a closed environment. As video surveillance systems growmore complex, the computational power distribution and data communication become a more important issue. Traditional centralized systems are no longer cope with the scale of processed data and inventive system designs are required to efficiently analyse the images at real time. The proposed system takes an hierarchical agent based approach to visual surveillance and distributes tasks over multiple intelligent cameras, while aiming for an equal distribution of computational demand of each camera. The tasks delegation applies an hierarchical model of agents and performs the analysis on the first layer of agents, which has all the required data available. The bottleneck of processing power demand and data stream within the global process is avoided by reducing the information resolution for each layer. The design concept of the hierarchical multi-agent approach to visual surveillance is tested in a simulated environment, which will only indicate the potential of object detection and tracking in real life environments. The main contribution of the research is to show the potential of applying the task delegation design method in distributed visual surveillance systems to achieve the scalability demand.