 The proposed method uses stereo vision to reconstruct unstructured scenes with a large variety of textures. It employs an optimized SAD fast-feature detection algorithm and a GVDS feature extraction algorithm to improve the overall efficiency and matching quality of the stereo matching stage. This allows for more accurate reconstructions and faster processing times compared to existing methods. The results of our experiments show that the proposed method has great potential for practical applications. This article was authored by Ming Jiu Chen, Zheng Xiu Duan, Zheng Xiaolan, and others.