Multi-Modal Detection and Mapping of Static and Dynamic Obstacles in Agriculture for Process Evaluation

Area of Science:

• Agricultural Engineering
• Robotics
• Computer Vision

Background:

• Autonomous agricultural vehicles require robust obstacle detection for safe operation.
• Existing systems need enhanced perception for full autonomy in complex field environments.

Purpose of the Study:

• To develop and evaluate a multi-modal obstacle and environment detection system for agricultural fields.
• To improve the safety and efficiency of autonomous agricultural vehicles through advanced perception.

Main Methods:

• Utilized a pipeline integrating range sensors (lidar, radar) and cameras (stereo, thermal) for obstacle detection.
• Implemented late fusion of sensor data into semantical occupancy grid maps for accurate mapping.
• Employed a Hidden Markov model for extracting process-specific parameters along the vehicle's trajectory.

Main Results:

• Multi-modal sensor fusion significantly enhances obstacle detection performance.
• Accurate traversability assessment and semantic mapping of field elements (crop, ground, obstacles) were achieved.
• The system effectively identifies unexpected structures, informing control systems for safer navigation.

Conclusions:

• Combining diverse sensor modalities is crucial for robust obstacle detection in agriculture.
• Tailored sensor fusion strategies are necessary for optimal performance across different detection tasks.
• The proposed approach advances the realization of fully autonomous agricultural vehicles.