Real-time straight line extraction and VLSI architecture

We present an efficient algorithm suitable for VLSI implementation for extracting the line segments from the input grey scale image. We first apply an edge detection algorithm to the input image producing an edge map and also recording for each pixel an orientation and confidence level. To allow a pipelined implementation we scan this map in a raster fashion. The next step uses the orientation and confidence level data to interpolate missed edge pixels. Then each line segment is traced in parallel, with a prioritisation scheme dealing with segment intersection and applying colinearity restrictions to avoid merging two segments of different slopes. Since the raster scan oversegments lines that form a small positive angle with the horizontal axis, so the scan is repeated from right to left and the two results are merged. We describe a video-rate architecture for hardware implementation of our algorithm