Study on Efficient Large Disparity Estimation for Objects at Short Distance

High-accuracy depth information in short distance scenes is essential for AR and MR especially of medical applications. Current mainstream deep learning-based disparity estimation does not consider enough for such scenes with large disparity of stereo vision. We propose a new disparity estimation method to handle large-disparity regions based on efficient construction of disparity cost volume. Our experimental results demonstrate its efficiency on higher accuracy of large disparity without additional computational cost.