A Systematic Solution for Moving-Target Detection and Tracking While Only Using a Monocular Camera.

Area of Science:

• Computer Vision
• Robotics
• Signal Processing

Background:

• Accurate 3D moving target detection and tracking is crucial for applications like autonomous systems and surveillance.
• Traditional methods often require multiple cameras or complex sensor setups.
• Existing 2D-based tracking systems face challenges in depth estimation and accuracy.

Purpose of the Study:

• To develop a robust visual target tracking system capable of 3D localization using only a 2D camera.
• To enhance the speed and accuracy of moving target detection in complex environments.
• To propose a novel geometrical approach for depth estimation from 2D measurements.

Main Methods:

• Utilizing an improved Pyramid, Warping, and Cost Volume Network (PWC-Net) for efficient optical flow-based detection.
• Employing a clustering algorithm to isolate moving targets from background noise.
• Implementing a geometrical pinhole imaging algorithm combined with a Cubature Kalman Filter (CKF) for 3D position estimation.

Main Results:

• The proposed system successfully detects and tracks moving targets in 3D space using a single 2D camera.
• The modified PWC-Net and clustering algorithm demonstrated effective target extraction from noisy data.
• The geometrical solution accurately calculated target azimuth, elevation, and depth with high computational efficiency.

Conclusions:

• The developed visual target tracking system offers an effective and computationally efficient solution for 3D moving target localization using only 2D visual input.
• The integration of advanced optical flow techniques with geometrical estimation provides a robust framework for real-world applications.
• The method's simplicity and speed make it suitable for real-time tracking scenarios.