Systems and Methods for Video-Based Fruit Weight Estimation Using Multi-Frame Image Geometry and Population-Derived Reference Data

Abstract

Abstract A computer-implemented system and method are provided for estimating fruit volume, weight, and / or yield from orchard video captured during row-wise scanning by a moving handheld or vehicle-mounted camera. A plurality of video frames corresponding to a same fruit item are selected, for example using a unique identifier assigned to the fruit item across the video. The frames represent different views of the fruit item as the camera moves relative to the fruit. A segmentation mask, such as a polygon mask, is generated for the fruit item in each selected frame, an ellipse is fitted to each mask, and observable minor-axis or semi-minor-axis values are extracted and aggregated to derive an image-observed minor-axis characterisation. A corresponding major-axis or semi-major-axis value is determined from population-derived characterising data obtained from laboratory measurements of representative fruit items. A volume estimate is calculated using an ellipsoidal or spheroidal model and converted to a weight estimate. Weight estimates may be combined with fruit counts to generate yield estimates, weight maps, heatmaps, or other spatial orchard outputs. 20 26 20 49 06 24 J un 2 02 6 2 0 2 6 2 0 4 9 0 6 2 4 J u n 2 0 2 6

Images