Abstract: Maize weeding robots play a significant role in the green production of modern agriculture. A high production efficiency can also be gained to reduce the labor costs, particularly with no chemical pollution. The autonomous operation of maize weeding robot can depend mainly on the accurate detection and tracking of maize seedling belt. In this study, an updated real-time recognition was proposed to extract the navigation line between the maize seedling belts using Region of Interest (ROI), in order to facilitate the seedling belt detection of maize weeding robot. A monocular camera was used to capture the maize seedling belt in front of the robot. Images preprocessing was firstly implemented to effectively segment the seedling zone and background using optimal threshold selection, according to the large gap between the green area of seedlings and weeds. Secondly, the feature points of seedling belt were accurately extracted using an improved adaptive response threshold Small univalue segment assimilating nucleus (SUSAN) corner method. The redundant outliers of feature points were then removed to reduce the amount of clustering calculation for the better real-time performance. Thirdly, the regional hierarchical clustering was incorporated into the sequential cluster, in order to improve the speed of clustering. The maize seedlings were also performed on the regional sequential clustering with clustering process. Moreover, the least square method (LSM) was used to fit each maize seedling belt. Finally, the ROI was adjusted to update the navigation line in real time, according to the heading deviation of the robot and the lateral deviation relative to the maize seedling belt. Meanwhile, the kinematics model was utilized to optimize the steering angle under the PID controller. The optimal steering angle was obtained to avoid the pressure and damage from the wheel skid using the seedling belt row tracking of the weeding robot. The navigation control system was developed using OpenCV library on the Visual Studio platform. A real-time detection of maize navigation line was realized for the synchronous interaction between the navigation line deviation under the STM32 control terminal, and the manual control of weeding robot navigation. A continuous video was randomly captured from the collected maize seedling belt data set as a test, in order to verify the accuracy of the real-time navigation line extraction. Results show the accuracy rate of navigation line extraction was 96.8% with the average processing time of 87.39 ms, indicating the excellent real-time and anti-interference. The average tracking error of maize seedlings with the straight lines and curves was less than or equal to 1.42cm, while the standard tracking error was less than or equal to 0.41cm in the simulated environment. In the farmland, the average tracking error of navigation at different speeds was less than or equal to 1.51cm, and the standard error was less than or equal to 0.44cm, indicating the accurate operating of the weeding robots in the maize seedling belt rows. In summary, the seedling belt extraction and navigation tracking control of weeding robot was precisely realized by the navigation control system of maize weeding robot. The fast and accurate identification of seedling band can be suitable for the strong adaptability and stability of navigation tracking control system, fully meeting the navigation requirements of maize weeding robot operation. The finding can provide the technical support for the subsequent research of weeding robots.