A noval weighted distance and local density double histogram used in 3D local feature description

In the field of computer vision, it is a challenging task to accurately describe local features in point clouds with different noise, resolution and low overlap rate. Weighted distance and local density double histogram (WDLDDH), an effective and robust 3D local feature description method, is proposed in this paper. There are two key ideas in the method. Firstly, secondary feature matching is introduced in the matching process to enhance matching accuracy. Secondly, two novel feature descriptors are proposed. One is weighted distance descriptor, the other is local sphere density descriptor. The former processes the distance between points in the local point cloud and the spatial marked points by using the method of “reciprocal weighted smoothing,” which makes the feature descriptor a high level of robustness. The latter adds an effective weight and increases the matching dimensions to correct the incorrect point pairs. The effectiveness of WDLDDH is verified using publicly available datasets, including the Stanford Repository and the SHOT project page. WDLDDH performs better in these commonly used accuracy evaluation metrics, such as Recall versus 1-Precision Curve (RPC), AUC values, and MSE. The experiments fully demonstrate that WDLDDH possesses superior stability and robustness.