Distinguishing Aerial Intruders from Trajectory Data: A Model-Based Hypothesis-Testing Approach

TL;DR

This paper presents a statistical hypothesis-testing algorithm to distinguish aerial intruders like birds and drones using velocity data, enabling efficient identification in UAS security systems.

Contribution

It introduces a model-based hypothesis-testing approach that simplifies detection to an explicit computation, improving computational efficiency and learning from archived data.

Findings

Explicit detector based on autocorrelation simplifies implementation.

Total probability of error is analytically characterized.

Simulations validate the effectiveness of the proposed method.

Abstract

Motivated by security needs in unmanned aerial system (UAS) operations, an algorithm for identifying airspace intruders (e.g., birds vs. drones) is developed. The algorithm is structured to use sensed intruder velocity data from Internet-of-Things platforms together with limited knowledge of physical models. The identification problem is posed as a statistical hypothesis testing or detection problem, wherein inertial feedback-controlled objects subject to stochastic actuation must be distinguished by speed data. The maximum a posteriori probability detector is obtained, and then is simplified to an explicit computation based on two points in the sample autocorrelation of the data. The simplified form allows computationally-friendly implementation of the algorithm, and simplified learning from archived data. Also, the total probability of error of the detector is computed and characterized. Simulations based on synthesized data are presented to illustrate and supplement the formal analyses.