High Speed Autonomous Navigation of Unmanned Aerial Vehicles using novel Road Identification, Following & Tracking (RIFT) Algorithm

Autonomous road navigation in unmanned aerial vehicles flying at very low altitudes paves the way for a multitude of applications such as security surveillance, monitoring of traffic or pollution, package delivery, ground- vehicle tracking etc. Most research in the field of autonomous road vehicles is focused on lane tracking and other marker- dependent techniques. In these lines, a novel, computationally efficient method for (front-view) road identification and tracking based on monocular vision is proposed which is designed to work on roads independent of markings. Road identification is implemented using a combination of spectral component analysis, edge energy-based filtering and morphological processing. Road identification and tracking works at 30 frames/sec for a frame size of 120 x 160 pixels, which is profiled on an Odroid XU4 mini-computer. Results indicating the performance of the proposed method as assessed on Cityscapes and KITTI datasets are presented.