Developing a Depth Estimation Algorithm for an Ultra-thin Light-Field Camera

In this study, we developed an ultra-thin multi-aperture light field camera system and algorithms for depth estimation. Depth estimation is a field that can be utilized broadly. It can be classified into monocular depth estimation, and stereo depth estimation. Generally, in order to acquire absolute-valued depth estimation, stereo depth estimation, based on the disparity informtion, is needed. For stereo depth estimation, we need to capture the scene using at least two cameras, with known baseline, then process the images to get the depth map. Therefore, we need a specific camera system for this task. In this research, the camera system, comprising 9 lenses with 3 different aperture sizes on an IMX477 sensor, captures multi-view images in a single shot. This setup enables the extraction of depth information using stereo matching techniques. Despite challenges related to noise and distortion due to the compact design and lack of an objective lens, our approach demonstrates the potential for accurate depth estimation. Future improvements in manufacturing and lens design are expected to enhance the image quality and depth estimation accuracy further.