Binocular Stereo Vision-based Navigation Information Research In Crop Fields

It’s a common knowledge that traditional agricultural machinery operation has large labor intensity, less affected by human factors, so that it’s an inevitable trend for manual operation replaced by intelligent, automatic agricultural machinery equipment in modern agriculture. And automatic navigation technology has become one of the primary problems to be solved in the area. Machine vision based navigation technology is becoming one of the hot researches in crop navigation because of its advantages in crop path identification, crop status extraction and other aspects. In which, the feature extraction and matching of crop plant is one of the key technologies for acquisition of crop row structure information in binocular vision.In this paper, the SIFT method is selected to conduct feature points extraction from the crop field images for its good stability based on the comparison and analysis of several feature extraction methods like Harris, Fast, SIFT, SURF and ORB. And, to promote the accuracy and efficiency of the feature matching of crop field images, Mask Image is proposed to filter feature points in the target crop area, retaining the feature points in target area and discarding the non-target ones. Averagely, the Mask Image can discard about2/3of total feature points those are non-crop in the experiment. At the same time, the optimized local area Epipolar-Constraint strategy is involved to fulfill the feature matching of binocular images. And the floating range of epipolar is set6pixels, which can satisfy the feature matching of cabbage and maize crop in the experiment. For the complexity of navigation of agricultural machinery in fields, the method of multi position calibration is adopted in the camera calibration process, and the distance between calibration board and camera is set2m,3m and5m respectively, and the height of camera is set70cm. The internal and external parameters of the binocular system are set as the average of multi position ones. Also, the extraction and matching of feature points, and navigation information structure statistics are carried out of the binocular images from different height (80cm/100cm) and different angle (10degree/30degree). The results of experiment show that the method of agricultural machinery equipment navigation based on binocular vision involved in the paper is correct and effective.