3D object recognition is a critical task in fields such as computer vision and pattern recognition. Recently, point cloud-based 3D object recognition techniques have been widely applied in areas such as automated grasping, remote sensing, navigation, and medical diagnostics. However, factors such as noise, occlusion, and repetitive patterns in the scene can lead to a large number of incorrect correspondences (mismatches) during the feature matching stage, and generate many erroneous pose transformation hypotheses, which affects the accuracy and efficiency of pose estimation and increases the difficulty of object recognition. To reduce the mismatch rate and improve recognition performance, this paper presents a 3D object recognition method based on an improved Hough voting approach. First, we introduce a mismatch removal algorithm based on point pair feature constraints (PPFC), which uses a matching score to filter high-quality matching subsets. Next, we propose an enhanced Hough voting hypothesis generation algorithm that effectively minimizes reliance on the local reference frame (LRF), allowing for efficient hypothesis generation using only a 2D Hough space. Additionally, we design a hypothesis validation algorithm based on an improved threshold strategy to assess and optimize the generated hypotheses. Experimental results show that our method achieves recognition rates of 100% on the Random Views dataset and 98.40% on the UWAOR dataset, demonstrating superior accuracy compared to existing 3D object recognition methods. This indicates that our approach can effectively identify objects even in the presence of significant occlusion and background interference. Moreover, our method offers high spatial and temporal efficiency, providing a novel and effective solution to the challenges of 3D object recognition.