|
Time-to-contact estimation in landing scenarios using feature scales| old_uid | 10215 |
|---|
| title | Time-to-contact estimation in landing scenarios using feature scales |
|---|
| start_date | 2011/09/29 |
|---|
| schedule | 13h45 |
|---|
| online | no |
|---|
| summary | I will present a feasibility study on using the scale of SIFT-features for estimating time-to-contact in space landing scenarios. The time-to-contact is a measure of vertical velocity divided by the height, and can in principle be used directly for controlling the lander. However, in this feasibility study the lander is assumed to have a radar altimeter, using the time-to-contact for estimating the vertical velocity. Experiments with virtual zooms on images show that : (1) the image size, the number of frames per second, and the amount of memory are important factors for the success of using feature scales for time-to-impact estimates, and (2) the distribution of estimates has many outliers, necessitating robust estimators such as the median. Experiments with landing scenarios in the PANGU simulator show that the vertical velocity estimates are accurate enough for successful landing in different conditions.
Bio-sketch : Guido de Croon obtained his M.Sc. degree in Artificial Intelligence at Maastricht University, performing the research for his thesis at the Consiglio Nazionale della Ricerca in Rome under supervision of Stefano Nolfi. Subsequently, he obtained his Ph.D. degree at Maastricht University on the topic of ``Adaptive Active Vision’’. His Ph.D. work involved using evolutionary algorithms for adapting local sampling models to various visual tasks. At the end of his Ph.D. period, he performed a half-year research visit to the laboratory of Dario Floreano at the École Polytechnique Fédérale de Lausanne, where he became enthusiastic for research on flying robots. Consequently, he joined the Micro Air Vehicle lab of the Technical University of Delft with the goal of developing computationally efficient computer vision algorithms for autonomous flight. Much of his work in Delft focused on algorithms for the DelFly II, a 16 g flapping wing MAV. Recently, he has joined the Advanced Concepts Team of the European Space Agency. There he researches biologically inspired algorithms for artificial intelligence in space. |
|---|
| responsibles | Fenouil |
|---|
| |
|