Autonomous vegetation identification for outdoor aerial navigation

Identification of landmarks for outdoor navigation is often performed using computationally expensive computer vision methods or via heavy and expensive multi-spectral and range sensors. Both choices are forbidden on Micro Aerial Vehicles (MAV) due to limited payload and computational power. However, an appropriate choice of the hardware sensor equipment allows the employment of mixed multi-spectral analysis and computer vision techniques to identify natural landmarks. In this work, we propose a low-cost low-weight camera array with appropriate optical filters to be exploited both as stereo camera and multi-spectral sensor. Through stereo vision and the Normalized Difference Vegetation Index (NDVI), we are able to classify the observed materials in the scene among several different classes, identify vegetation and water bodies and provide measurements of their relative bearing and distance from the robot. A handheld prototype of this camera array is tested in outdoor environment.