Navigation line extraction method for combine harvester under low contrast conditions

Abstract: In grain harvesting , the combine harvester's header need to guide along the harvest boundary in time to ensure the harvester is working in full widt, which requires the operator has high operational skills. In addition, with the long-time operation the driver is prone to fatigue, which brings safety hazards to agricultural production. Therefore, it is of great significance to study the automatic navigation technology of combine harvester to reduce the labor intensity of drivers and improve production efficiency. The key of automatic navigation is the extraction of navigation line. Due to richer environmental information, wider detection range and more complete information, visual-based navigation methods have attracted extensive attention. However, the contrast between cut and uncut areas of mature wheat in the image is extremely low under strong illumination, which leads that the harvest boundary of crop is quite blurred. To solve the problem that the cut edge is difficult to extract under low contrast conditions, a fast and accurate navigation line extraction method of combine harvester based on region growing algorithm is presented in this paper. Firstly, the color images of harvesting scene collected by camera was converted into the gray scale image, and the Gaussian filtering was applied to remove the image noise. Then, the region growing algorithm was used to segment the image. Initial seed was selected based on some criteria and then the uncut wheat area was segmented by region growing process. The gray value of each 4-neighboring pixel was compared with the mean gray value of the seed region, if their difference was smaller than the threshold the corresponding pixel was added to the seed region. This procedure was repeated until no pixel could be grouped in the region. To improve the robustness of the region growing algorithm, an adaptive threshold selection method based on gray histogram was proposed. The multi-peak Gaussian fitting of the gray histogram was performed and half of the absolute value of the difference between mean values of the two Gaussian components was taken as the threshold of region growing, then the segmented binary image was processed by the morphological operations to fill the small holes in the segmented region which made the harvest boundary of wheat smoother. Finally, the harvest boundary of crop was detected and the harvester navigation line was acquired by fitting the harvest boundary with the least squares method. The experimental results showed that even though the contrast of cut and uncut wheat areas was extremely low, the proposed method could accurately detect the wheat harvest boundary and extract the harvester navigation line. Under different operating conditions such as different light intensity and different crop growth density, the average angle error between the navigation line extracted by the proposed method and the manually calibrated navigation line was less than 1.21 °, processing a 900×1 200 pixels image took about 0.4 s, which basically meets the real-time requirements of the combine harvester navigation. The results can provide a reference for the automatic navigation of the combine harvester.