Pose estimation has garnered significant attention in recent years and has found extensive application in fields such as autonomous driving, robotics, and augmented reality. In the current research, point cloud recognition algorithms based on point-pair-features have been shown to be effective in recognizing objects and pose estimation, but redundant points included in the characterization of model features degrade the recognition performance and computational efficiency of the algorithms. To address this issue, this paper introduces curvature features to filter out unnecessary points and enhance the expression of model features. The resulting global model description is stored in a hash table, and the estimated pose is obtained through the combination of curvature-weighted voting and the Iterative Closest Point (ICP) algorithm for optimization. Additionally, a background removal technique is proposed for fixed usage scenarios, which significantly improves operational efficiency in real-world situations. Experimental results using various datasets and real environments demonstrate that the proposed approach reduces redundancy, improves point-pair feature (PPF) expression, and enhances recognition rate and matching speed by 4.7% and 46.7%, respectively.