Aiming at solving the the problems of excessive time consumption and low registration efficiency caused by the4PCS algorithm when registrating the point cloud data, a improved4PCS coarse registration method based on the3D-SIFT feature point was proposed. The point cloud data of the cherry tree was collected from four directions by DK depth camera. Firstly, a point cloud denoising framework was designed by using traight-through filtering and statistical filtering to screen high-quality three-dimensional point cloud. Secondly, the SIFT algorithm was applied to extract features from cherry tree point cloud, which reduced data dimensions and enhanced feature stability. Thirdly, the obtained set of points about source feature and target feature were used as initial data of the 4PCS algorithm, and the coarse registration was carried out. Finally, after obtaining the precise pose, the ICP algorithm was used for precision registration until the best matching state was achieved. Taking cherry tree point cloud data of different types as the experimental objects to registration experiments, the time consuming and the root maen square error indexes were introduced to evaluate the experiments. In the coarse registration stage, the results showed that the registration time of the proposed registration method was 4.16 s and 4.33 s, respectively. The root mean square error was 0.953 cm and 1.810 cm, respectively, which effectively reduced the registration error and shortened the registration time. The results of multiple precision registration experiments demonstrated that both the overall point cloud registration time and registration error achieved optimal values based on the fusion of the proposed method and the ICP algorithm in the precision registration. The whole registration time was 4.84 s and the root mean square error was 0.845 cm.