High Performance Customizable Architecture for Machine Vision Applications

Vision based applications are present anywhere. A special market is industry, allowing to improve product quality and to reduce manufacturing costs. The vision systems applied to industries are known as machine vision systems. These systems must meet time constraints to operate in real time. Generally the production lines are more and more fasters, and the time to process and bring a response is minimal. For this reasons, dedicated architectures are emplaced. In this work a review of several commercial systems is presented, as well a proposed architecture is depicted. The architecture is concern as a customizable platform, avoiding having knowledge in hardware description languages. It is based on massive parallelism to achieve the maximum processing performance. Several optimizations at different levels are applied to increase the final system speedup. Also, time and area metrics are reported, showing that the architecture is well suitable for real time video processing in industrial applications.
Facultad de Informática