Research on the Relative Position Detection Method between Orchard Robots and Fruit Tree Rows.

Area of Science:

• Agricultural Robotics
• Computer Vision
• Machine Learning

Background:

• Accurate relative positioning of orchard robots to fruit tree rows is crucial for autonomous navigation.
• Existing methods for estimating robot-to-row position parameters suffer from low accuracy.

Purpose of the Study:

• To propose a novel machine vision-based method for detecting the relative position of orchard robots and fruit tree rows.
• To improve the accuracy of position parameter estimation for autonomous orchard robot navigation.

Main Methods:

• Utilized an improved YOLOv4 model for fruit tree trunk identification.
• Employed binocular camera triangulation to calculate tree trunk coordinates and coordinate conversion for ground projection.
• Applied linear fitting with the least squares method to determine the navigation path and robot position parameters.

Main Results:

• The improved YOLOv4 model demonstrated a 5.92% increase in average accuracy and a 7.91% increase in average recall rate for fruit tree trunk detection compared to the original YOLOv4.
• The proposed method achieved average errors of 0.57° for heading angle and 0.02 m for lateral deviation.
• Accurate calculation of heading angle and lateral deviation was validated across different inter-row positions.

Conclusions:

• The developed machine vision method significantly enhances the accuracy of orchard robot positioning.
• This approach provides a reliable reference for autonomous visual navigation systems in orchards.
• The improved YOLOv4 model offers superior performance in fruit tree detection for robotic applications.