Aiming at the problems of uneven points, indistinct feature points and broken feature lines in point cloud feature line extraction, a feature extraction method based on sector projection and sliding window connection is proposed. Firstly, the global density of the point cloud is analyzed and the average point spacing is calculated； then, the sector projection method is used to count the distribution of adjacent points in the sector, and the boundary points of the point cloud model is determined from the distribution； in turn, the local triangle normal is calculated, sharp edge point is judged according to the normal mutation, the local probe line is extended on the basis of the sector projection, and the relevant strategy for judging the non-sharp edge point is formulated according to different probe line types； finally, based on the sliding window connection method, the unordered feature points are connected in an orderly manner to obtain smooth feature line. The experiments show that the method can extract the characteristic lines of the model relatively completely, and has a certain anti noise ability at the same time. Compared with the other two methods, the method in this paper is more sensitive to the weak features in the model and the extraction effect is relatively better.