| Most of the traditional agriculture work is labor-intensive and dependent on labor experience strongly. Intelligent navigation of agricultural machinery will be one of effective ways to solve the problem.Navigation based on binocular vision has received more and more attention because of low cost, rich information and so on.Firstly, feather extracting algorithms are used to extract feature point, and then different feature point descriptors are selected according to different feature points. We analyze effect of different constraints, and then the boundary constraint rule is proposed base on the characteristic of the fileld vision navigation system. The paper obtains the binary image first and then designs the baseline at the middle of the binary image. The density curve is obtained by scanning a sector region on the baseline. Furthermore, the density, the width and the angle of the crop row can be computed conviently. Other crop rows are obtained by designing Sector-Scan density model and angle constraints. Then logistic regression is applied to recognize the nearest crop. The experiment shows that the accuracies of the first crop and the second crop are 97.7% and 94.7%respectively over five sets with different crops and backgrounds. The elevation map of the crop is obtained by fusing the disparity and the coordinates of plant feature points and then the elevation map is used to filter those points that we do not need in the experiments. The enhanced elevation map is obtained by extending feature point height to the neighborhood. The enhanced elevation map and binary image are put together to get the confidence image. The sector-scan method can search probable lines over the confidence image. Some search tables and the polymerization are designed to reduce the processing time. The row processing time is 0.3s and the standard deviation is 0.004s. Under laborary conditions, the experiment shows that the max deviation is not lager than 10cm under laborary conditions. |
PDF Full Text Request