6DOF Pose Estimation of a 3D Rigid Object Based on an Adaptive Model Curvature Point Pair Features

Currently, point cloud recognition algorithms that utilize the point pair feature (PPF) have demonstrated superior matching capabilities compared to their competitors even under conditions of occlusion and noise. However, including redundant point pairs in the global model description greatly diminishes the algorithm's efficiency. Insignificant scene reference points can readily result in incorrect matches. Additionally, different models require distinct parameter settings, making it challenging for new objects. Therefore, this paper focuses on the PPF algorithm and presents a 6D pose estimation method using an adaptive model curvature of Point Pair Features (AMC-PPF). The AMC-PPF algorithm is based on the PPF algorithm. Moreover, by utilizing adaptive truncation curvature, we reduce the unnecessary global model point pairs and select scene reference points that provide more geometric information. Furthermore, we adjust the sampling interval of scene reference points to maintain a consistent number of reference points, thereby improving accuracy. To verify the effectiveness of our method, we analyze the experimental results on the public dataset. The results demonstrate that the algorithm reduces redundant point pairs, it improves algorithm efficiency while ensuring algorithm accuracy.