A dynamically reconfigurable architecture for vision pre-processor

The main challenge in implementation a vision system is to design a vision chip that has high computing capability and flexibility, simple programmability and operation, in addition to high throughput. In this paper, a new architecture of a dynamically reconfigurable processor is proposed. With a reconfigurable convolution array module (RCAM), the processor can finish the visual pre-processing tasks based on convolution in parallel. One of the salient novel features in this processor is the hardware configure module (HCM). By using it, users can reconfigure the processor easily without understanding the hardware details. The paper describes the architecture and performance. We demonstrate that the proposed processor can deal with 720P video format in real time.