Model-Based Signal Processing for Joint Drones Detection, Tracking, and Parameters Estimation

Abstract

The problem of multicopter (multirotor drone) classification is considered. A two-stage approach for multicopter joint detection, tracking, and parameter estimation is proposed. A previously published particle filter (PF)-based track-before-detect (TBD) algorithm with a single-rotor drone is used in the first stage to detect, localize, and track the target. The algorithm is, however, modified by the utilization of a new drone model, which is based on a simplified representation of a rotated propeller as a bunch of thin wires. Using this model, closed-form analytical equations for the radar signal temporal dependence and micro-Doppler spectrum are derived for each rotor. Significant improvement in micro-Doppler spectrum prediction due to the implementation of this model has been observed. The actual number of multicopter rotors and their independent parameters, such as rotation velocity and initial orientation angle, are estimated in the second processing stage. The estimation problem is formulated as a maximum likelihood (ML) search in a multidimensional space of parameters. This computationally expensive optimization problem is converted to the problem of multiple likelihood function peaks detection in 2-D space “rotational velocity-initial orientation angle” for each propeller. The latter is solved by a computationally efficient 2-D grid search algorithm, which is followed by a few extra processing steps to remove the residual false alarms by analyzing detections over multiple consecutive coherence processing intervals. The proposed approach for multicopter detection and classification has been verified using simulated and experimental data.