Strawberry Plant Localization via Relative Pixels in Sequential Images

Abstract

When conducting precision operations, such as disease detection, weed removal, yield prediction, and harvesting, on plants such as strawberries and blueberries, it is necessary to know the exact location of each plant. To date, GPS and LiDAR based methods have been proposed, however these methods either cannot routinely store position data, are labor intensive, expensive, or bulky. In this study, a low cost and lightweight localization approach is proposed using relative pixel information of adjacent plants. The kinematic information of a scouting robot carrying the camera and the relative position information of adjacent plants are modeled. The centroids of strawberry plants are identified one by one via image processing technologies. An extended Kalman filter is then developed to estimate the relative positions of adjacent plants. The proposed strawberry plant localization algorithm is validated in a commercial farm. The method is low cost and can be used in routine localization operations in agricultural fields.