A method to track two-dimensional objects through a sequence of images is presented. The scene is supposed to be rigid while the camera is moving, and objects have to lay in a plane perpendicular to the optical axis of the camera, being constant the distance from the camera to this plane. A Hough Transform-like technique is used to select matches consistent with the rigidity constraint. Then, motion parameters are found through a least-squares minimization. The motion found is used to match the rest of the objects. A real application to match plants in a crop field is presented, it uses sequences of outdoor images taken under natural illumination. Images are segmented and the resulting regions are input to the tracking system. Before the tracking method is applied, regions undergo a clustering process to identify individual plants. Shape descriptors invariant to rotations and translations are computed for each plant, and they handle the splitting and merging problem of regions. Results using real images are presented.