The position accuracy is mainly decided by the accuracy of landmark directional angle for the localization system of agricultural machinery based on omnidirectional vision sensor. The algorithms about low pass filter eliminating noise and artificial landmark features detection was studied. After eliminating noise, the red pixels intensity was calculated and the red landmark pixels beyond the threshold were extracted. Based on the Euclid geometrical distance of neighbor pixels, the small area of pixels was divided and the center of gravity was calculated for the extracted small area representing the candidate of one landmark. The algorithm provided the blue patch as compensation to further distinguish the landmark from other objects in a complex environment. The landmark detection test proved that the low pass filter eliminating noise was effective. Pointing, straightaway driving and tilt tests were conducted on the level ground in a 30 m×30 m area outside. Results showed that the average values of MAE and RMSE of landmark direction angles were 1.12° and 2.10°, and the influence of vehicle vibration and tilt angle less than 5° were not obvious. The positioning accuracy was about 10 cm. The proposed algorithm is feasible, the error of landmark direction angle is less and the program is fast running speed and robustness. 