1 / 11

A Machine Learning Approach to the Visual Perception of Forest Trails for Mobile Robots

A Machine Learning Approach to the Visual Perception of Forest Trails for Mobile Robots. A. Giusti , J. Guzzi , D. Ciresan , F. Lin He , J. Pablo Rodríguez F. Fontana, M. Faessler , C. Forster J. Schmidhuber , G. Di Caro, D. Scaramuzza , L. Gambardella

migueljohnson
Télécharger la présentation

A Machine Learning Approach to the Visual Perception of Forest Trails for Mobile Robots

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. A Machine Learning Approach to the Visual Perception of Forest Trails for Mobile Robots A. Giusti, J. Guzzi, D. Ciresan, F. LinHe, J. Pablo RodríguezF. Fontana, M. Faessler, C. ForsterJ. Schmidhuber, G. Di Caro, D. Scaramuzza, L. Gambardella IDSIA, USI-SUPSI, Lugano, SwitzerlandRPG, Universityof Zurich, Switzerland

  2. Steering a MAV with a purely reactive controller

  3. Our focus Perception of the trail direction Mapping Obstacle perceptionand avoidance Path Planning Visual Odometry Control

  4. A challenging pattern recognition problem Trail heading left Trail heading straight ahead Trail heading right

  5. An image classification problem P(trail is left) Raw pixel values Input P(trail is straight) Deep Neural Network P(trail is right)

  6. Internal view of the classifier

  7. How the classifier was trained

  8. Dataset size

  9. Classification accuracy Our classifier (accuracy: 85%) Human observer (accuracy: 86%)

  10. Application to video acquired from an handheld camera

  11. Conclusions • Pure machine learning approach to trail perception • Performance comparable to humans when tested on individual frames of the test set • Onboard odroid U3 runs SVO pipeline and DNN at more than 15FPS

More Related