Agile wide-field imaging with selective high resolution.

Area of Science:

• Optics and Photonics
• Computer Vision
• Aerospace Engineering

Background:

• Wide-field and high-resolution (HR) imaging are crucial for applications like aviation reconnaissance and safety monitoring.
• Current methods demand large detector arrays, leading to high complexity and cost.
• Natural scenes often exhibit statistical sparsity, with important targets concentrated in small regions of interest (ROI).

Purpose of the Study:

• To develop an agile wide-field imaging framework with selective high resolution using minimal hardware.
• To enable real-time identification and tracking of ROIs for continuous HR imaging.
• To reduce the cost and complexity associated with traditional HR imaging systems.

Main Methods:

• Utilizing a dual-camera system: one short-focal for wide-field low-resolution imaging and one long-focal for ROI HR imaging.
• Implementing a deep-learning-based multiscale registration algorithm for automatic, real-time ROI localization.
• Employing a gimbal-mounted long-focal camera for continuous ROI tracking.
• Leveraging the statistical sparsity prior of natural scenes to focus on relevant areas.

Main Results:

• Demonstrated a proof-of-concept imaging setup weighing only 1181 grams.
• Successfully integrated the system onto an unmanned aerial vehicle (UAV) for air-to-ground monitoring.
• Achieved a 120° wide field of view (FOV) with a selective 0.45mrad instantaneous FOV for HR imaging.

Conclusions:

• The proposed framework offers an efficient and cost-effective solution for wide-field, selective HR imaging.
• The deep-learning registration method is robust to significant setting variations between cameras.
• This agile imaging system has practical applications in aerial surveillance and monitoring.