Crop edge detection based on stereo vision

This paper focuses on the development of a crop edge detection algorithm based on the point cloud produced by a stereo camera system using the GPU for fast matching of the camera images. The approach utilizes the 3D characteristics of the transition between the crop and the stubbles or the ground. Therefore, the point cloud is sorted into a grid of cells to create an elevation map. A segmentation in crop and ground is obtained using the Expectation-Maximization algorithm with a Gaussian Mixture Model to represent the distribution of the cell's heights. This segmentation is Bayesian filtered over a short time frame to create a more robust segmentation result. Afterward, the resulting potential crop edge locations are processed using robust linear regression to come up with an overall linear crop edge model. The implemented system has been tested in a series of experiments with detailed results stated at the end of this work. (C) 2019 Elsevier B.V. All rights reserved.