Abstract: Aiming at the complex greenhouse environment where the ground is bumpy and the branches and leaves block the road, the research on the inter-rows navigation method of greenhouse robot based on the fusion of camera and LiDAR data was carried out. Firstly, the improved U-Net model was used to realize the accurate and fast segmentation of image road area. Secondly, the ground point cloud was pre-segmented by fusing the image segmentation result to reduce the incline of the point cloud data caused by the ground bumpiness. Then, the improved KMeans algorithm was used to realize the rapid clustering of the crop row point cloud, and the cluster centers were used as the main area points of crop rows to reduce the influence of branches and leaves blocking the road on extraction of crop row centerline. Finally, the RANSAC algorithm was used to fit the crop row equations on both sides and calculate the navigation lines. The navigation line accuracy was evaluated by experiment, the validation work was conducted in two greenhouse scenarios at three typical greenhouse robot operation speeds. The experimental results showed that the performance and timing of the segmented images met the requirements of subsequent point cloud pre-segmentation; the experiment of point cloud data frames by bumpy environment can effectively calibrate the ground point cloud; compared with the raster height difference segmentation of ground point cloud, the segmentation effect was better and the time consumption of single frame processing was increased very little; on the test set, more than 94% of the data frames can accurately extract the navigation line and the average angle error was not higher than 1.45°. The research result can meet the greenhouse robot along the crop row autonomous navigation driving requirements.