Determination of the number of green apples on trees using color imaging

An algorithm for estimating the number of apples in color images acquired in orchards is presented. The algorithm includes four main steps: (1) Detection of pixels that have a high probability of belonging to apples; (2) Formation and extension of "seed areas", which are connected sets of pixels that have a high probability of belonging to apples; (3) Segmentation of the contours of these seed areas into arcs and linear segments; and (4) Combination of these arcs and comparison of the resulting circle with a simplistic "apple" model. The performance of the algorithm is investigated using two datasets. For the first datasets, which consists of images recorded in full automatic mode of the camera and under various lighting conditions, more than 85% of the apples are correctly detected but direct illumination and color saturation cause a large number of false positive detections. For the second dataset, which consists of images that were manually underexposed and recorded under mostly diffusive light, close to 90% of the apples are detected while the false positive detection rate is less than 5%.