研究目的
To propose a new algorithm for estimating an accurate disparity map in stereo matching, addressing sensitivity to low texture areas, noise, and variations in brightness and contrast.
研究成果
The proposed stereo matching algorithm combining SAD and GF produces accurate disparity maps, particularly in low texture regions, and performs competitively with established algorithms on the Middlebury dataset, demonstrating effectiveness in noise reduction and edge preservation.
研究不足
The algorithm is a local-based method, which may have lower accuracy compared to global methods in some scenarios; computational efficiency and performance on other datasets are not extensively discussed.
1：Experimental Design and Method Selection:
The algorithm uses Sum of Absolute Differences (SAD) for matching cost computation, Guided Filter (GF) for cost aggregation and post-processing, and Winner-Takes-All (WTA) strategy for disparity optimization.
2：Sample Selection and Data Sources:
The Middlebury stereo benchmarking dataset with 15 training images is used for evaluation.
3：List of Experimental Equipment and Materials:
A computer system with Windows 10,
4：2GHz processor, and 8GB memory. Experimental Procedures and Operational Workflow:
Steps include SAD computation, cost aggregation with GF, disparity selection with WTA, left-right consistency checking for occlusion handling, fill-in process for invalid pixels, and final noise removal with GF.
5：Data Analysis Methods:
Quantitative evaluation based on bad pixel percentage for all pixels and non-occluded pixels using the Middlebury benchmarking system.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
