DroBoost: An Intelligent Score and Model Boosting Method for Drone Detection

TL;DR

This work tackles robust drone detection under challenging conditions, where drones appear small and amid clutter. It advances a YOLOv5-based detector by fusing a diverse real-plus-synthetic data strategy guided by model error analysis, a binary drone/non-drone classifier to recalibrate confidences, and a Kalman-based track scoring mechanism to exploit temporal information. The key contributions include offline-rendered synthetic data targeting FP/FN cases, a separate binary classification stage with a calibrated confidence update, and a nuanced track-scoring scheme that stabilizes predictions over time. The approach achieves state-of-the-art performance on Drone-vs-Bird benchmarks, including a 1st Place finish at ICIAP 2021, and demonstrates improved robustness and explainability for real-time drone detection.

Abstract

Drone detection is a challenging object detection task where visibility conditions and quality of the images may be unfavorable, and detections might become difficult due to complex backgrounds, small visible objects, and hard to distinguish objects. Both provide high confidence for drone detections, and eliminating false detections requires efficient algorithms and approaches. Our previous work, which uses YOLOv5, uses both real and synthetic data and a Kalman-based tracker to track the detections and increase their confidence using temporal information. Our current work improves on the previous approach by combining several improvements. We used a more diverse dataset combining multiple sources and combined with synthetic samples chosen from a large synthetic dataset based on the error analysis of the base model. Also, to obtain more resilient confidence scores for objects, we introduced a classification component that discriminates whether the object is a drone or not. Finally, we developed a more advanced scoring algorithm for object tracking that we use to adjust localization confidence. Furthermore, the proposed technique won 1st Place in the Drone vs. Bird Challenge (Workshop on Small-Drone Surveillance, Detection and Counteraction Techniques at ICIAP 2021).

Figures (1)

• Figure 1: Generated synthetic data similar to examples where the real model performs poorly. Synthetic data similar to FP predictions above (a), synthetic data similar to FN predictions below (b).