Exploring the Potential of High-Resolution Drone Imagery for Improved 3D Human Avatar Reconstruction: A Comparative Study with Mobile Images

Reconstructing a 3D avatar of a human body from 2D images is not an easy task since it depends on several processing steps to enhance 3D acquisition devices input. Using camera-equipped drones with computer vision algorithms and photogrammetry tools to capture high-quality imagery data for 3D models reconstruction has become widespread recently. In this paper, a comparative study between 3D human avatars reconstruction based on 2D images captured from two different photography sources is conducted. Each acquisition device with an integrated lens has the same resolution parameters: a DJI Mavic Mini2 drone with a 12-megapixel camera lens and an iPhone 8 Plus with a 12-megapixel camera lens. To collect 2D images of a group of volunteers simulating a crowded human scene in both instances, the lens is approximately 5 m away from the human model. We used BUFF dataset to train our model and produce a high-quality 3D reconstructed human avatar. To validate our work, we employed the following evaluation metrics: PSNR, MSE, and SSIM. The obtained results showed that reconstructing 3D avatars from 2D images captured by a drone is more sophisticated in terms of quality and clarity. In contrast, images captured by a smartphone suffers from noise and distortions.