<p>Traditional pose estimation algorithms often require manual adjustment of various thresholds for parameter settings in order to enhance algorithm performance, making it difficult to meet the needs of unmanned and intelligent production. Therefore, this paper focuses on addressing the above issues arising from preprocessing and pose estimation during the grasping process. First, in the preprocessing stage after acquiring the point cloud, an adaptive radius filtering algorithm is introduced to tackle the problems of high algorithmic complexity and poor adaptability in point cloud denoising. To address the challenge of setting distinct registration thresholds for point clouds with varying quantities or types, an improved PCA + ICP algorithm with adaptive thresholding is proposed for robust pose estimation and target registration. The effectiveness of the proposed algorithm is demonstrated through experiments on both public datasets and in-lab collected data, achieving pose estimation accuracies of 0.038 cm and 0.031 cm, recognition accuracy of 0.98, and a processing time of 3.1 s. Moreover, when facing occluded point cloud registration, the algorithm achieved a MAD of 0.027 m and an IQR of 0.056 m, showing strong overall performance. Consequently, the method fulfills the timing constraints of real-world industrial processes, all while maintaining strong performance in terms of precision and stability.</p>