Architectures for Computer Vision: From Algorithm to Chip with Verilog

4 Verilog Vision Simulator

In designing a vision chip, testing is more than simply checking for the correct relationship between the actual output and the desired output, given the test vectors, as in the case of ordinary digital circuit design. The vision chip must deal with very large amounts of data and high data rates such as video frames, as well as various image formats such as bitmap, jpg, and tiff. The data rate in a stereo vision chip is expected to be two to three times higher than it is in an ordinary single video system. Moreover, the vision chip might be part of a large vision system that consists of various types of software and hardware. To build such a vision system, we have to interconnect two heterogeneous systems: vision software and a vision chip. Therefore, we have to move away from the conventional test bench concept to the more specialized concept of the vision simulator. This vision system will function as an interface to the chip and vision programs, in a similar fashion to open source computer vision (OpenCV). Generally, the simulator will perform a long sequence of operations: provide images to the chip, perform preprocessing, supply the intermediate preprocessed data to the chip, retrieve the result from the chip, perform post-processing, and provide the final output. In addition to simulation, the simulator by itself can be used as a research tool to develop vision algorithms. Thus, with a simulator, we can build a complete vision system comprising ...