The contour features of ancient ceramic fragments can directly affect the quality and efficiency of the restoration of cultural relics. In order to solve the problems of low precision and time-consuming in contour extraction due to the thin body, irregular shape and large amount of point cloud data of ancient ceramic fragments, a contour extraction algorithm of ancient ceramic fragments based on the density of neighborhood point set is proposed. Firstly, the Oriented Bounding Box (OBB) center plane parallel cutting plane is used to slice the fragments for realizing the layered processing and data simplification of the point cloud. Secondly, according to the law that the density of neighborhood points at contour points and non-contour points is different, the density feature of neighborhood point set is combined with the Random Sampling Consensus (RANSAC) algorithm to achieve accurate and fast extraction of fragment contours. Finally, construct the space classification plane and classify the fracture surface and non-fracture surface contours based on spatial positional constraints. The experimental results show that the running time of the algorithm can be controlled within 15~25 seconds, and the accuracy of contour extraction can reach 78.3%, with high accuracy and integrity, which can provide technical basis for digital restoration of ancient ceramic cultural relics.