Repeatability of edge detectors in various environmental conditions

Abstract

This paper explores the interactions between environmental conditions and edge detector performance for use as a critical function for detect and avoidance (DAA) operations in the unmanned aerial systems (UAS) industry. The goal of this study was to establish the best edge detection scheme for sunny, low light, cloudy, and foggy conditions. A laboratory test chamber was developed to simulate these conditions. Canny, Laplacian of Gaussian (LoG), Roberts, and Prewitt edge detectors were evaluated. Images were taken in these conditions and Pratt's Figure of Merit was used to evaluate the repeatability of the edge detector. It was determined that Prewitt performs best in low light and cloudy conditions, while LoG performs well in foggy conditions. Prewitt was shown to be the best overall edge detector with the properly chosen threshold. These results were validated with a natural image to show laboratory produced images can be used in place of images taken outdoors for this study. This validation proved laboratory settings produce comparable results to real world conditions.