Depth map improvements for stereo-based depth cameras on drones

Abstract

Using stereo-based depth cameras outdoors on drones can lead to challenging situations for stereo algorithms calculating a depth map. A false depth value indicating an object close to the drone can confuse obstacle avoidance algorithms and lead to erratic behavior during the drone flight. We analyze the encountered issues from real-world tests together with practical solutions including a post-processing method to modify depth maps against outliers with wring depth values.