Abstract
This article presents a monocular vision-based algorithm for detecting obstacles and identifying obstacle-aware regions, developed to be used for collision avoidance by a multicopter. The first step of our algorithm is to predict a disparity image from a single-view image via implementing a deep encoder-decoder network. All pixels in this disparity prediction are then categorized as one of three classes, obstacle, road, or obstacle-free, by combining V-disparity analysis and a fuzzy inference system. For pixels belonging to obstacle objects, obstacle-aware regions are generated within the field of visual perception. To accommodate the safety margins of a multicopter, intermediate waypoints are then added to obtain a new flyable path that passes through an unknown environment safely. Experimental results verified the effectiveness of the detection of obstacles and the identification of obstacle-aware regions. The accuracy of disparity prediction and monocular depth estimation were quantitatively compared to support the feasibility of monocular vision in obstacle avoidance. Furthermore, the entire algorithm was successfully tested on a robotic platform, autonomously flying a hexacopter in an outdoor space with obstacles. In conclusion, the proposed monocular algorithm performs well for obstacle detection and depth estimation and is potentially an alternative to a binocular solution.
Highlights
Over the past decade, the use of unmanned aerial vehicles (UAVs) has rapidly grown, in non-military purposes
We present a computation-efficient approach for classifying each pixel in the disparity image into three classes: road, obstacle, and obstacle-free
By accumulating the computation time of each step, the framerate is about 8 to 10 frames per second (FPS) which we deem acceptable to guide a multicopter in general applications
Summary
The use of unmanned aerial vehicles (UAVs) has rapidly grown, in non-military purposes. Godard’s approach has been extended by involving trinocular assumptions [19], and utilizing semantic information, as discussed in [20] and [21] Another form of self-supervision is to employ monocular video sequences, in which the consecutive image frames play the role of supervisory signals. Surveying learning-based monocular depth estimation methodologies over the past few years, the Monodepth model presented in [17] outperformed the majority of previous works extant at that time and motivated several subsequent studies. Based on the monocular disparity or depth (inverse to each other) estimation, we attempted to improve the Monodepth model and apply the result to avoid collisions while navigating a small multicopter.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
