Depth estimation based on defocus blur using a single image taken by a tilted lens optics camera

Abstract

In various fields such as measurement of the distance between cars and three-dimensional recognition in robotics, the depth needs to be estimated over a wide range. For this purpose, we propose a depth estimation method that uses a ground-in-focus image taken by a tilted lens optics camera. Our proposed method detects edges by local scale control. It then measures the blur amount at each edge by fitting a Gaussian blur model to them. If there are other neighboring edges, we use the linear sum of the Gaussians to increase the fitting accuracy with the blur model. We can then obtain a depth map based on the blur amount and y-coordinate of the pixel. Our experimental results showed the effectiveness of our proposed method.