Stereo Vision Based Trajectory Tracking for Agricultural Vehicles

Abstract

The purpose of this research was to develop a stereo vision based trajectory tracking method for agricultural vehicles in open field environments. This paper reports the outcomes from the first phase of research, which focused on trajectory tracking in a relatively flat agricultural field based on detected relative motion of natural point features in consecutively acquired images. A binocular stereo camera mounted in front of an agricultural utility vehicle was used to collect those stereo images of the field. A method based on Harris corner detection algorithm was developed to find those feature points. A multiscale iterative feature tracking method was then applied to match the identified feature points in consecutive image frames. Such matched feature points were used to generate motion vectors for tracking the vehicle traveling trajectory. To represent the tracked trajectory in vehicle independent format, it is necessary to convert the trajectory tracked in image coordinates to vehicle coordinates. To create an image-based coordinates transformation method, the plane field images captured by the left and right lenses were also formed a series of disparity images of the field being observed. Those disparity images were used to generate 3D views of the field for providing the necessary information in formulating the image-based algorithm of coordinates transformation. Field tests validated that the created methods could track the trajectory of an agricultural vehicle traveling in an open agricultural field with a relative error about 6% in distance estimation.