摘要:
In our developed localization system for auto navigation based on omnidirectional vision, we used artificial landmarks. In order to distinguish landmarks from natural environments effectively in a bigger application area, this paper analyzes the correlations between the landmark image size and landmark position, landmark height and camera height. Then to propose a landmark model with the right circular cone made of red and blue patches and a fast landmark tracking algorithm based on color intensity difference. Outdoor experiments were conducted with the bottom diameter of landmark was 25 cm and the height about 100 cm under natural sunlight in a 50x50-m square area to verify the artificial landmark model and tracking algorithm. Experimental results show that proposed landmark model is feasible and landmark features to be detected effectively. The RMS error of x-axis, y-axis and distance are 29.75 cm, 17.32 cm and 34.24 cm, respectively. The artificial landmark design and tracking algorithm are applicable for localization system.
摘要:
This paper introduces an artificial landmark self-localization method using omnidirectional vision for agricultural vehicles field road navigation. We propose a landmark model and the algorithm to track landmark and calculate the absolute location of camera based on omnidirectional image. Red and blue landmark pixels beyond the threshold were extracted as a small area and the center of gravity of the extracted small area was calculated representing the landmark position candidates. The distance between landmark and camera in the image was estimated and transformed the image distance to spatial distance using distance computational model. The absolute location of camera was calculated by circle theorems. Outdoor experiments were conducted on a flat asphalt road in the field under natural sunlight. Experimental results showed that the RMS and mean distance errors are less than 24 cm in a 20 m distance. The self-localization method is feasible.