Efficient Sparse to Dense Stereo Matching Technique

Abstract

Acquiring accurate dense depth maps with low computational complexity is crucial for real-time applications that require 3D reconstruction. The current sensors capable of generating dense maps are expensive and bulky, while compact low-cost sensors can only generate the sparse map measurements reliably. To overcome this predicament, we propose an efficient stereo analysis algorithm that constructs a dense disparity map from the sparse measurements. Our approach generates a dense disparity map with low computational complexity using local methods. The algorithm has much less computation time than the existing dense stereo matching techniques and has a high visual accuracy. Experiments results performed on KITTI and Middlebury datasets show that our algorithm has much less running time while providing accurate disparity maps.