Design of Crop Yield Estimation System for Apple Orchards Using Computer Vision

Crop yield estimation is an essential element in apple orchard management. Apple growers currently predict yield based on historical records and manual counting. These methods require extensive experience on the part of farm managers to take into account variations in weather, soil conditions, pests, etc., and are generally labor-intensive and inaccurate. In this work, we propose an automatic computer vision system for detecting and counting red apples to predict crop yield in orchards. The system is composed of a low-cost two-camera rig with ring flashes, and custom computer vision algorithms to process images. The camera rig is mounted on an automated utility vehicle. It acquires images of trees at nighttime using the flashes to avoid unpredictable variations in natural lighting conditions. Our algorithms use color as a visual cue to detect red apples from a tree. Apple detection results and geographic coordinates of the vision system (obtained by GPS) are used to locate and count the fruit. We present here the current system configuration, and its preliminary evaluation on a dataset collected at the Sunrise Orchard in Rock Island, WA, in September 2011.