Mapping the variability of fruit size and quality within tree canopies in commercial orchards is an important tool for implementing precision horticulture. To do so at a reasonably fast rate requires localization technologies that offer sufficient speed and accuracy, at a range long enough to cover entire trees – or several trees at a time. Existing approaches for measuring fruit locations include: manual (centimeter accuracy and measurement time in the order of minutes per fruit); interacting magnetic fields (centimeter accuracy and range limited to 2-3 meters); and computer vision (centimeter accuracy, short measurement time; calibration and non-occluded fruit requirements).

This paper presents the design, error characterization and testing of a mobile system which utilizes radio waves and trilateration to measure the locations of fruits inside tree canopies. The components and measurement methodology are briefly described next.

Range measurement between two points is performed by the PulsON 400 Ranging and Communications Modules (P400 RCM) by Time Domain, Inc., USA. The P400 RCM is an Ultra Wideband (UWB) radio transceiver that measures the distance between two P400 RCM antennas. A mobile trailer uses four of these units as ‘beacons’ and a fruit picker carries one unit on his belt; the unit is connected to an antenna on his glove. The time instant when the picker grasps a fruit is recorded by clicking a button on an Microsoft XBOX controller. At that time instant, the four distances from the trailer beacons to the picker’s hand are also recorded. By combining the four distances of the glove antenna from the trailer beacons, the three-dimensional coordinates of each glove are computed, with respect to the trailer. The trailer is equipped with a high-precision Real-time-kinematic GPS that measured the geographical position of the trailer and an inclinometer, i.e., an attitude sensor that measured roll, pitch and heading trailer angles. Using the data from these two sensors, the geo-referenced coordinates of the worker glove (fruit position) with respect to the world frame (UTM) can be computed.

After the system components were built and the data acquisition software was completed, the error histograms of the P400 RCM ranging units in free space and foliage were assessed experimentally. Based on these error distributions, a parametric optimization was performed to calculate the positions for the trailer beacons that would minimize the overall positioning error. The system achieved accuracy better than 20 cm, 95% of the time (mean error is 11 cm) within a large digitizing volume of 15 m³ and a fruit position digitization rate of approximately 1 fruit per second. The system was tested successfully in commercial orchards. Its range was big enough to cover one or two large open-vase trees or several smaller trellised trees in commercial Californian orchards. More trees in a row were covered by moving the trailer along the row. The system proved practical for measuring fruit locations in commercial orchards and its use can be extended to trees of various architectures and training systems.