Path extraction algorithm of orchard visual navigation based on scene dynamic weighted fusion

Abstract

<b><sc>Abstract.</sc></b> Aiming at the problems that visual navigation system faces complex image background and many interference factors in orchard environment, this paper proposed an orchard navigation path extraction algorithm based on scene dynamic weighted fusion. Firstly, the orchard image was pitch collected by a camera on rack, and the original image was converted to HSV space. The image was preliminarily segmented according to predefined threshold. After median filtering and morphological processing, the image was divided into three scenes: sky, canopy and soil. Secondly, connected domain segmentation algorithm was used to extract sky and soil scenes accurately, and Canny operator was used to detect and filter scene edges. Thirdly, the scene coordinate points were selected discretely, and the visual reference lines of sky and soil were fitted by least square method. The intersection points of sky and soil image reference lines were calculated when the pitch angle is equal. The sky reference line was mirror projected by intersection to realize the unification of coordinate axis with soil reference line. Finally, weighted fusion of sky and soil detection phase reference lines. constructed the deviation confidence to extract final navigation path. In this study, the Camellia oleifera orchard was selected as experimental environment. The pitch camera was used to shoot the orchard video, and 150 frames of images were intercepted to test the algorithm. The experimental result shows that the algorithm was used to extract the visual navigation path without prior knowledge. The extraction accuracy is 86.3%, and the single frame image processing time is about 45ms, which meets the requirements of real-time and robustness of orchard interline navigation.