Fruit detectability analysis for different camera positions in sweet-pepper.

Area of Science:

• Agricultural Robotics
• Computer Vision
• Horticultural Engineering

Background:

• Robotic harvesting of sweet peppers requires accurate fruit detection and localization.
• Plant structures often cause partial fruit occlusion in single-viewpoint images, hindering robotic systems.

Purpose of the Study:

• To investigate the impact of multiple camera positions and viewing angles on sweet pepper fruit visibility and detectability.
• To determine optimal camera configurations for enhancing fruit detection in greenhouse environments.

Main Methods:

• Developed a recording device to capture images from various azimuth, zenith, and horizontal positions along crop rows.
• Manually assessed fruit visibility and determined fruit detectability (FD) for 14 distinct camera positions.
• Evaluated the combined effect of multiple camera viewpoints on overall fruit detectability.

Main Results:

• Single-viewpoint imaging with max 50% occlusion yielded a fruit detectability (FD) of no more than 69%.
• Optimal single positions involved front views with a 60° zenith angle looking upwards.
• Combining five selected camera positions increased the maximum FD to 90%.

Conclusions:

• Multiple camera viewpoints are crucial for overcoming occlusion issues in robotic sweet pepper harvesting.
• Strategic camera placement and multi-view fusion significantly enhance fruit detection accuracy.
• This research provides a foundation for developing more robust sensor systems for automated agricultural tasks.