A novel algorithm for distant aircraft detection

Abstract

In this paper a novel algorithm for distant aircraft detection for visual sense-and-avoid for UAV is presented. The algorithm uses local edge density to partition the frame into two types of regions. The first type is the unstructured or homogeneous part like sky region and the second part where there is a structured background, like high contrast clouds or terrain regions. The airplanes are detected on the two types of regions with different strategies. The algorithm was planned to run in an embedded environment with low power consumption, thus it can be run onboard of a small or mid-size UAV. First steps towards the GPU implementation on the nVidia Jeston TK1 development board are done and also presented in the paper.

Detection of aircraft below the horizon for vision‐based detect and avoid in unmanned aircraft systems

“…where target looming information in the sky region is extracted from HMM filter outputs. As an alternative solution, Zarandy et al, recently suggested that the sensor aircraft could execute regular heading changes to improve observability of targets with minimal motion with respect to the ground clutter. Future work will investigate the suitability of our system for detecting aircraft with minimal motion relative to the background ground clutter.…”

“…Importantly, Zarandy et al . recently noted that true collision threats can appear stationary with respect to ground clutter [as illustrated in Fig.…”

“…For small‐to‐medium sized fixed‐wing UAS, vision is the leading sensing technology for detecting aircraft that are uncooperative (in the sense that they do not share information) . While vision‐based techniques for reliably detecting aircraft against sky backgrounds have been experimentally demonstrated, only partial progress has been made in the detection of aircraft against structured backgrounds such as the ground clutter in Figure . In this paper, we propose a vision‐based system for reliably detecting potential collision threats that appear below the horizon (against ground clutter).…”

“…Several groups have recently made partial progress in developing vision‐based systems for detecting potential collision threats when they appear below the horizon against structured ground clutter . These systems have predominantly used frame differencing techniques to detect collision threats that have apparent motion with respect to background ground clutter.…”

“…These systems have predominantly used frame differencing techniques to detect collision threats that have apparent motion with respect to background ground clutter. Although these techniques have had their initial feasibility demonstrated, detection systems based on frame differencing may fail during true collision encounters when collision threats have small apparent motion between consecutive frames . Furthermore, frame differencing is often insufficient for rejecting ground clutter, and so current state‐of‐the‐art systems in Nussberger et al., also use digital terrain maps.…”