Real-world UAV image deblurring by PSF calibration

Unmanned aerial vehicles (UAVs) have been increasingly employed in a wide range of applications, including environmental monitoring, precision agriculture, and infrastructure inspection. However, image blur frequently occurs in UAV-acquired imagery, compromising visual quality and adversely affecting high-level vision tasks such as object detection, classification, and 3D reconstruction. This study addresses image blur originating from camera-related factors. Without access to the optical design specifications, we estimate the point spread function (PSF) across various fields of view through camera calibration. These PSFs are subsequently utilized to synthesize a blurred dataset for training a deep neural network, which is then employed to restore sharpness in real-world captured images. Additionally, we assess the performance of several existing deblurring techniques and compare them with our PSF-driven approach. Experimental results indicate that our method outperforms competing techniques in real image deblurring scenarios.