Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Estimation and Tracking of a Moving Target by Unmanned Aerial Vehicles

Jun-Ming Li, Ching Wen Chen, Teng-Hu Cheng

TL;DR

The paper addresses monocular UAV tracking of a moving target with unknown dynamics and intermittent visual observations. It combines an unscented Kalman filter with a constant-velocity target model and online process-noise covariance adaptation to estimate target position and velocity from YOLO bounding-box measurements, even when detections are temporarily unavailable. The estimated target velocity is injected as a feedforward term into an image-based visual servoing controller, which uses a reduced interaction matrix to compute the camera commands and keep the target in view. Simulations in ROS/Gazebo validate the approach, showing improved tracking performance and convergence relative to static-noise models and highlighting practical considerations such as detector outages and multi-angle detection challenges.

Abstract

An image-based control strategy along with estimation of target motion is developed to track dynamic targets without motion constraints. To the best of our knowledge, this is the first work that utilizes a bounding box as image features for tracking control and estimation of dynamic target without motion constraint. The features generated from a You-Only-Look-Once (YOLO) deep neural network can relax the assumption of continuous availability of the feature points in most literature and minimize the gap for applications. The challenges are that the motion pattern of the target is unknown and modeling its dynamics is infeasible. To resolve these issues, the dynamics of the target is modeled by a constant-velocity model and is employed as a process model in the unscented Kalman filter (UKF), but process noise is uncertain and sensitive to system instability. To ensure convergence of the estimate error, the noise covariance matrix is estimated according to history data within a moving window. The estimated motion from the UKF is implemented as a feedforward term in the developed controller, so that tracking performance is enhanced. Simulations are demonstrated to verify the efficacy of the developed estimator and controller.

Estimation and Tracking of a Moving Target by Unmanned Aerial Vehicles

TL;DR

The paper addresses monocular UAV tracking of a moving target with unknown dynamics and intermittent visual observations. It combines an unscented Kalman filter with a constant-velocity target model and online process-noise covariance adaptation to estimate target position and velocity from YOLO bounding-box measurements, even when detections are temporarily unavailable. The estimated target velocity is injected as a feedforward term into an image-based visual servoing controller, which uses a reduced interaction matrix to compute the camera commands and keep the target in view. Simulations in ROS/Gazebo validate the approach, showing improved tracking performance and convergence relative to static-noise models and highlighting practical considerations such as detector outages and multi-angle detection challenges.

Abstract

An image-based control strategy along with estimation of target motion is developed to track dynamic targets without motion constraints. To the best of our knowledge, this is the first work that utilizes a bounding box as image features for tracking control and estimation of dynamic target without motion constraint. The features generated from a You-Only-Look-Once (YOLO) deep neural network can relax the assumption of continuous availability of the feature points in most literature and minimize the gap for applications. The challenges are that the motion pattern of the target is unknown and modeling its dynamics is infeasible. To resolve these issues, the dynamics of the target is modeled by a constant-velocity model and is employed as a process model in the unscented Kalman filter (UKF), but process noise is uncertain and sensitive to system instability. To ensure convergence of the estimate error, the noise covariance matrix is estimated according to history data within a moving window. The estimated motion from the UKF is implemented as a feedforward term in the developed controller, so that tracking performance is enhanced. Simulations are demonstrated to verify the efficacy of the developed estimator and controller.

Paper Structure

This paper contains 15 sections, 20 equations, 6 figures.

Table of Contents

  1. Introduction
  2. Preliminaries and Problem Statement
  3. Kinematics Model
  4. Image Model
  5. Measurement Model
  6. Position and Velocity Estimation
  7. Unscented Kalman Filter
  8. Estimation of Noise Covariance Matrices
  9. Tracking Control
  10. Target Recognition
  11. Controller
  12. Simulations
  13. Environment Setup
  14. Simulation Results
  15. Conclusion

Figures (6)

  • Figure 1: Kinematics model.
  • Figure 2: The images of the dynamic targets are captured from an onboard camera on the multirotor in the Gazebo simulator. Note that the center of the bounding box is considered as a feature point for the subsequent analysis, and the bounding boxes, enclosing the vehicles from different angles of inclination, are obtained from a YOLO network that is trained for this work.
  • Figure 3: Block diagram of the controller.
  • Figure 4: Estimate and ground truth of the target trajectory in the $XY$ plane in the global frame.
  • Figure 5: Estimate and ground truth of the target velocity in the $X$ and $Y$ direction of the global frame.
  • ...and 1 more figures

Theorems & Definitions (4)

  • Remark 1
  • Remark 2
  • Remark 3
  • Remark 4