In order to improve the autonomy and safety of orchard spray robot walking among orchard rows, a fruit tree location method based on 3D LiDAR and optimized DBSCAN algorithm was proposed. Firstly, the three-dimensional LiDAR was used to obtain the environmental information of the orchard in real time, and the original data was preprocessed by region of interest extraction, ground point cloud segmentation and voxel filtering. Then, the DBSCAN algorithm was optimized, the KD tree index was constructed to order the real-time point cloud data, and the KD tree nearest neighbor search was used to replace the traversal search method of the traditional DBSCAN algorithm. Finally, the clustering results of fruit trees were marked with reference position. Taking the midpoint of the x-axis of the plane between the rows where the clustering results faced as reference point, the point to the xoy plane of the chassis height of the orchard spray robot was projected, and the coordinates of the positioning reference point of the fruit trees were obtained, so as to calculate the relative position between the orchard spray robot and the fruit trees. The experimental results showed that compared with the traditional DBSCAN algorithm, the accuracy and real-time performance of the optimized DBSCAN algorithm were significantly improved. Based on the optimized DBSCAN algorithm, the average horizontal positioning error of fruit trees was 2.6%, and the average vertical positioning error of fruit trees was 1.6%. When the orchard spray robot traveled between rows, this method can meet the accuracy and real-time requirements of fruit tree positioning, and it can provide effective reference for precision agriculture equipment in autonomous navigation and operation in forest orchard environment. 