Abstract: Aiming at low-coverage-rate point clouds, a registration algorithm was proposed based on
local features in the paper. Firstly, local features including the local depth, deviation angle between
normals and point cloud density are extracted, and the local feature descriptor is obtained. Secondly,
the correspondence of local feature sets is calculated and the corresponding candidates are gained.
Thirdly, the outliers are eliminated and coarse registration is achieved. Lastly, an improved iterative
closest point (ICP) algorithm based on the rotation angle constraint, and the dynamic iterative
coefficient is employed to complete fine point cloud registration. The experiment results reveal that
the point cloud registration algorithm could achieve the precise registration of low-coverage-rate
point cloud, based on local features, a high-precision and fast one.

Key words: point cloud registration, local feature, iterative closest point, rotation angle constraint;
dynamic iterative coefficient