Visual navigation path extraction of orchard hard pavement based on scanning method and neural network

Abstract

The accurate extraction of navigation path is very important for the automatic navigation of agricultural robots. Aiming at the complex orchard environment and the problem that the existing navigation path extraction algorithms are too complex and narrow application range, a visual navigation path extraction method based on neural network and pixel scanning was proposed in this paper. This method trained the semantic segmentation network based on Segnet and Unet on the basis of orchard road condition data sets. According to the edge of the orchard road condition mask area, the navigation path was fitted by the designed scanning method, filtering algorithm and weighted average method. The experimental results showed that the segmentation accuracy of neural network under low light, ordinary light and strong light was 96.00%, 92.00% and 92.00% respectively. The average pixel error was 9.5 pixel and the average distance error was 5.03 cm. In the actual orchard environment, the orchard road was generally 3.0 m, the average distance error accounted for 1.67%. Therefore, this method improves the accuracy of orchard visual navigation path extraction, meeting the operation requirements of tracked robots in orchard, and provides an effective reference for visual navigation task.