The objective of VIEWS is to demonstrate the feasibility of a computer vision system for real-time surveillance of outdoor scenes which comprise moving objects in a known, wide-area and structured environment.
The issues involved in vision based robotics systems have been addressed. It was concluded that complete autonomy, although desirable, was not practical within current restrictions. The methodology of design was discussed and it was found that common to all methodologies was intuition based upon experience in the field.
A system was developed which was capable of inspecting a 300 mm by 280 mm circuit board containing 360 components in 50 seconds and 200 components in 39 seconds, averaging 7 components per second. The system proved far more reliable than manual inspection procedures.
A skeleton system for actively maintaining an interpretation of a scene has been developed. It uses rule based recognition procedures triggered by constantly active demons, thus permitting event detection and identification.
When contextual or semantic information was available, or some interpretation of the scene was required, it was found that a combination of knowledge based and computational systems was necessary. The combined system was able to exploit artificial intelligence tools and control techniques for error recovery.
Model based tracking was used to develop a system for vehicle monitoring and proved adequate under various conditions of lighting or weather. The system relies on a prior knowledge of the objects it will encounter.
VIEWS addresses R&D issues in the field of computer vision related to situation representation and reasoning and real-time control of various asynchronously arriving data-streams. VIEWS will also include the provision of support tools for acquiring the scene and object knowledge.
The VIEWS system prototype will be integrated into a generic architecture using a number of different components developed by the project partners, and will be demonstrated in two areas of application:
- surveillance of ground traffic in airports, and in particular, surveillance of aircraft servicing vehicles on apron areas at airports, for efficient aircraft turnround
- surveillance of vehicle traffic on roads and motorways for traffic control purposes.
These demonstrations will be developed and assessed in close cooperation with potential end-users (airport and road authorities).
This project will demonstrate the feasibility of computer vision for surveillance of moving objects in a structured and known environment. With rapidly decreasing costs for components, and the technical solutions obtained in this project, these vision systems are expected to be ready for market development for a range of surveillance applications.
92084 Paris La Défense
LE3 1UF Leicester
RG6 2AR Reading