DNDADI: density-weighted normal angle deviation and depth information local descriptors for 3D rigid data feature matching

3D local shape feature description plays a key role in the field of computer vision due to its ability to accurately recognize and match complex 3D objects, while the effects of noise, different mesh resolutions, clutter, and occlusion can reduce the distinguishing ability of the features and increase the risk of false matching. To address the above problems, this study proposes a new feature descriptor named Density weighted normal deviation Angle and depth information (DNDADI) composite descriptor based on density weighted normal deviation Angle and depth information (DNDADI). DNDADI adopts an innovative local reference frame (LRF). Local reference frame (LRF) design, which enhances the robustness of the LRF by estimating the density of the neighborhood points around the critical points when constructing the LRF. In addition, DNDADI creates a comprehensive, efficient and unique 3D local surface descriptor by combining the local normal angle deviation information of the key points in the LRF with the local depth information captured from the three orthogonal viewsheds and encoding it in an integrated manner. Rigorous performance evaluations on the B3R and UWAOR public datasets have shown that DNDADI exhibits superior results in model matching scenarios such as 3D shape retrieval and object recognition compared to FPFH, SHOT, TOLDI, and SDASS methods, and furthermore, DNDADI has been adapted to Geiger-Mode Avalanche Photodiode(GM-APD) LIDAR acquisition in a 3D object matching task, AUC_pr can reach 0.8832 with a matching time of 0.063s, further confirming its effectiveness.