FPGA implementation of contour detection based on Helmholtz principle

摘要

In recent years, research and development on advanced driver assistance systems and self-driving functions have been actively conducted in automobile industries. Contour detection of white lines on roads is one of the most fundamental tasks for such technologies and demand for its inexpensive implementation using a general camera has created various algorithms. In this paper, in order to detect contours of complex road white lines including curves with low latency and low power consumption, we discuss FPGA-friendly implementation of Furushou's algorithm which is based on Helmholtz principle: the human visual system does not perceive any structure in a random image. The evaluation results showed that the implementation achieved a latency of a few milliseconds and improved the performance by up to 3046 times compared to an embedded CPU implementation. The estimated power consumption was about 200 mW, which is considered to be applicable for small robots.