Alessandro Giusti¹, Jerome Guzzi¹, Dan Ciresan¹, Fang-Lin He¹, Juan Pablo Rodríguez Gómez¹, Gianni Di Caro¹, Jürgen Schmidhuber¹, Flavio Fontana², Matthias Faessler², Christian Forster², Davide Scaramuzza², Luca M. Gambardella¹
¹ Dalle Molle Institute for Artificial Intelligence (IDSIA), USI/SUPSI, Lugano Switzerland
² Robotics and Perception Group, University of Zurich, Switzerland
This work was supported by the Swiss National Science Foundation (SNSF) through the National Centre of Competence in Research (NCCR) Robotics, and the Supervised Deep/Recurrent Nets Grant (project code 140399).
We study the problem of perceiving forest or mountain trails from a single monocular image acquired from the viewpoint of a robot traveling on the trail itself. Previous literature focused on trail segmentation, and used low-level features such as image saliency or appearance contrast; we argue that such features are unsuitable in many real-world cases, and propose a different approach based on a Deep Neural Network used as a supervised image classifier. By operating on the whole image at once, our system computes the main direction of the trail compared to the viewing direction. Qualitative and quantitative results computed on a large real-world dataset show that our approach outperforms alternatives, and yields an accuracy comparable to the accuracy of humans which are tested on the same task.
A. Giusti et al.: "A Machine Learning Approach to the Visual Perception of Forest Trails for Mobile Robots". IEEE Robotics and Automation Letters (2016) and ICRA 2016 (accepted) IEEE Xplore link.
Previously: oral presentation at IROS 2015 Workshop on Vision-based Control and Navigation of Small, Lightweight UAVs.
@article{giusti2016machine,
title={A Machine Learning Approach to Visual Perception of Forest Trails for Mobile Robots},
author={Giusti, Alessandro and Guzzi, Jerome and Ciresan, Dan and He, Fang-Lin and Rodriguez, Juan Pablo and Fontana, Flavio and Faessler, Matthias and Forster, Christian and Schmidhuber, Jurgen and Di Caro, Gianni and Scaramuzza, Davide and Gambardella, Luca},
journal = "IEEE Robotics and Automation Letters",
year = "2016"
}Zenodo record with download links for the whole acquired dataset (training and testing, 15+ GB)
Download Video 1
Download Video 2
Download Video 3
Download Video 4