On-Tree Mango Fruit Size Estimation Using RGB-D Images.

Area of Science:

• Agricultural Engineering
• Computer Vision
• Horticulture

Background:

• In-field mango fruit sizing is crucial for estimating maturation, guiding harvest logistics, and marketing strategies.
• Accurate fruit size assessment from images requires precise estimation of camera-to-fruit distance.
• Existing machine vision methods primarily focus on fruit counting, not detailed size analysis.

Purpose of the Study:

• To evaluate low-cost technologies for in-field camera-to-fruit distance estimation.
• To develop and validate a machine vision system for rapid, in-field mango fruit size estimation.
• To determine the practicality and limitations of RGB-D cameras for this application.

Main Methods:

• Assessed RGB-D, stereo vision, and Time of Flight (ToF) cameras for distance measurement.
• Calibrated an RGB-D camera, integrating depth data with RGB images.
• Employed cascade detection with Histogram of Oriented Gradients (HOG), Otsu's method, and CIE L*a*b* color space for fruit detection and background removal.
• Utilized a 1D filter for pedicle removal and ellipse fitting for fruit identification.
• Calculated fruit dimensions using the thin lens formula with RGB-D depth data and image size.

Main Results:

• The RGB-D camera demonstrated good performance and cost-effectiveness, though it was sensitive to direct sunlight.
• The developed system achieved a Root Mean Square Error (RMSE) of 4.9 mm for length and 4.3 mm for width.
• Manual measurements showed a standard deviation of 1.2 mm, indicating the system's accuracy relative to human variability.

Conclusions:

• The RGB-D camera-based machine vision system offers a practical and user-friendly solution for in-field mango fruit sizing.
• The system's primary limitation is its reduced performance in intense direct sunlight.
• This work represents a significant advancement in practical, cost-effective machine vision for field-based fruit sizing.