Obstacle detection in snow covered terrain

Abstract

This work describes an automatic detection method of obstacles covered by snow. The method is based on the detection of statistical anomalies relative to an estimated background image which contains no obstacles. The sensitivity of the detection can be adjusted by a specified probability of false alarms, and the obstacle detection confidence is characterized by a probability of detection. Statistical properties of the background image are estimated from the given image without additional information on the background. The visible height of obstacles above the snow is related to the actual height of the obstacles above the ground, so that an operator can estimate the actual height of the snow covered obstacle. The developed method requires no training, is self-calibrating to the cluttered images, operates with a single given image, and aligns with a detection quantification adopted in the receiver operating characteristic framework.