Motion-guided small MAV detection in complex and non-planar scenes

TL;DR

This paper addresses the challenge of detecting extremely small MAVs from a moving platform in complex, non-planar scenes by proposing MGMD, a motion-guided detector that fuses motion cues, trajectory filtering, and appearance-based refinement. The method comprises three modules: motion feature enhancement to reveal moving MAVs, trajectory filtering to suppress parallax-induced false positives, and local fine detection combining a lightweight appearance classifier with a cropped-region detector. On the ARD-MAV dataset, MGMD significantly outperforms state-of-the-art methods and runs at 28 FPS, demonstrating robust performance for tiny targets in dynamic backgrounds. The work highlights the value of integrating pixel-level motion features with temporal tracking and localized appearance cues to achieve reliable small-object MAV detection in challenging aerial environments.

Abstract

In recent years, there has been a growing interest in the visual detection of micro aerial vehicles (MAVs) due to its importance in numerous applications. However, the existing methods based on either appearance or motion features encounter difficulties when the background is complex or the MAV is too small. In this paper, we propose a novel motion-guided MAV detector that can accurately identify small MAVs in complex and non-planar scenes. This detector first exploits a motion feature enhancement module to capture the motion features of small MAVs. Then it uses multi-object tracking and trajectory filtering to eliminate false positives caused by motion parallax. Finally, an appearance-based classifier and an appearance-based detector that operates on the cropped regions are used to achieve precise detection results. Our proposed method can effectively and efficiently detect extremely small MAVs from dynamic and complex backgrounds because it aggregates pixel-level motion features and eliminates false positives based on the motion and appearance features of MAVs. Experiments on the ARD-MAV dataset demonstrate that the proposed method could achieve high performance in small MAV detection under challenging conditions and outperform other state-of-the-art methods across various metrics

Figures (6)

• Figure 1: Examples of challenging conditions for MAV detection. In air-to-air MAV detection scenarios, the target MAV is often engulfed by a complex background or may be extremely small in the image. The yellow boxes enclose MAVs. (Top right images are a better view with 300$\%$ zoom-in).
• Figure 2: Testing results of the dense optical flow of moving MAVs extracted by the pre-trained model of the RAFT 2020RAFT. Yellow box indicates the MAV target. Black box indicates the extracted dense optical flow.
• Figure 3: The architecture of the proposed MGMD algorithm. First, we use frame alignment and three-frame difference to segment moving objects from complex backgrounds. Then, multi-object tracking is applied to generate trajectories for each detection, and a trajectory-based filter is used to eliminate false positives. Finally, a local appearance-based classifier and a local appearance-based detector are utilized on a cropped region to obtain precise detection results.
• Figure 4: An example of trajectory filtering. After the trajectory filtering, most of the bounding boxes generated by image alignment errors are removed.
• Figure 5: The qualitative comparison between the detection results from TPH-YOLOv5l, MEGA, and MGMD. MGMD can detect the target MAV when the appearance features are unreliable under challenging conditions. Yellow box indicates the target detected by our proposed method. Red box indicates the target detected by TPH-YOLOv5l. Blue box indicates the target detected by MEGA. Because the image size is too large compared to the target's size, we only show the cropped regions from the full-size image for a better view.