Self-calibration and Refocusing with One Interpolation for Infrared Array Cameras

Refocusing array camera images enables the adjustment of the focus area to bring objects on a specified plane into clear view while intentionally blurring surrounding regions. Traditional techniques involve re-interpolating images using calibrated parameters specific to the array camera. Nevertheless, the current calibration algorithms for infrared array cameras heavily lean towards laboratory calibration using checkerboard patterns, resulting in subpar performance in practical settings due to parameter variations. Additionally, the existing scene-based self-calibration methods for array cameras tend to overlook the pose invariance of sub-views, leading to misalignments caused by small disparities within each set. To tackle this challenge, we introduce a novel scene-based self-calibration approach for infrared array cameras. Our method involves: (a) capturing dual array frames to construct a light field with enhanced depth resolution; (b) adaptively selecting initial image pairs for superior reconstruction quality; and (c) employing pose self-calibration based on the pose invariance of sub-views, utilizing the EPnP algorithm for precise calibration. This method continually updates camera parameters, facilitating real-time, high-precision refocusing with minimal need for interpolation.