Robust Stereo Matching Based on Cost Volume Fusion for Optimal Disparity Estimation

In a stereo-based disparity estimation algorithm, matching cost evaluation is the key process to construct an initial cost volume. In practice, however, various challenges including occlusion region, reflection and radiometric changes result in inaccurately evaluated costs. This paper presents a local stereo matching method based on cost volume fusion. The proposed method consists of four steps: i) image transformation based on log-chromaticity, ii) cost volume generation, iii) cost volume reconstruction based on fusion, and iv) cost aggregation and optimization for optimal disparity estimation. The proposed method simply selects an optimal disparity map by combining various cost volumes with different properties. Therefore, the proposed method can be applied to various three-dimensional imaging applications, such as robot vision and autonomous vehicular systems.