Efficient Terrain Following for a Micro Aerial Vehicle with Ultra-Wide Stereo Cameras
| Autor: | Ingo von Bargen, Samantha Stoneman, Wolfgang Stürzl, Florian Steidle, Marcus Gerhard Müller |
|---|---|
| Jazyk: | angličtina |
| Rok vydání: | 2020 |
| Předmět: |
geography
geography.geographical_feature_category Stereo cameras business.industry Computer science Measure (physics) Terrain Autonomie und Fernprogrammierung Perzeption und Kognition Sample (graphics) Institut für Robotik und Mechatronik (ab 2013) Task (computing) Cave autonomous aerial vehicles cameras collision avoidance height measurement image sensors mobile robots path planning remotely operated vehicles robot vision stereo image processing Nadir Computer vision Artificial intelligence Aerospace business Stereo camera |
| Popis: | In recent years, Micro Aerial Vehicles (MAVs) have drawn attention to the aerospace community. With such autonomous flying platforms, it is possible to explore foreign extraterrestrial bodies in an efficient and faster manner than other robotic platforms, like rovers. In addition, they can be equipped with a variety of different sensors. Cameras are especially well suited, since they are light, energy-efficient and deliver a broad spectrum of information. Following the underlying terrain in a defined height is a fundamental task for any exploring MAV. To accomplish this, many systems possess a designated height sensor, which in most cases only delivers a single height estimation taken from nadir. In such a setup, the MAV is just adjusting its height based on the current height estimation and does not take any terrain lying ahead into account, which results in delayed height adjustments. In this paper, we propose a novel method based on a wide-angle stereo camera setup, which is attached to the MAV, to overcome such problems. Due to the wide vertical field of view, the vehicle is able to not only measure its current height, but also the terrain lying ahead. Therefore, the MAV is able to perform a better terrain following compared to other methods, which use just a single nadir height sample. Our algorithm only needs to take the depth image, calculated by the stereo cameras, and the estimated gravity vector into account. Therefore, our method is very fast and computationally efficient, compared to other methods, which build up an entire map beforehand. As a result, the procedure presented here is also suitable for tiny flying systems with low computational capabilities and memory resources. The terrain following algorithm runs in real-time and on board the system, and is therefore also suitable for confined environments, like caves, and where communication delays are present. We evaluate our method with simulated data and real tests on an MAV. To demonstrate that our method works in a variety of different terrains, we show experiments with different slopes and obstacles in the flight path. We also compare our method to a basic terrain following by using just a single height measurement in a more classical approach. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|