Shortwave infrared fusion for snow surface traversability mapping

Abstract

Estimating the mechanical properties of snow from imagery is an essential part of over-snow vehicle autonomy. However, snow surfaces that differ widely in strength, traction, and motion resistance tend to appear a uniform bright white in visible or broadband infrared imagery, and it is difficult to determine where an oversnow vehicle should operate from imagery alone. In this work we determine the optimal fusion of filtered broadband shortwave infrared (SWIR) imagery to separate snow types with different mechanical properties by appearance. We demonstrate vastly increased discrimination skill in distinguishing snow types using a small number of SWIR cameras in both field and laboratory settings, and also identify sources of environmental context that can improve lookahead sensing for oversnow vehicles. Overall, we show that a small set of inexpensive SWIR filters is a powerful tool for over-snow autonomy and motion planning.