Individual honey bee tracking in a beehive environment using deep learning and Kalman filter
Panadda Kongsilp1, Unchalisa Taetragool1, Orawan Duangphakdee2
1Department of Computer Engineering, King Mongkut's University of Technology Thonburi, Bangkok, 10140, Thailand.
Scientific Reports
|January 11, 2024
Summary
This study developed a deep learning and Kalman filter system to track individual honey bees, overcoming challenges like occlusion and background similarity. The system achieved high accuracy in segmenting and tracking bee behavior for ecological studies.
Area of Science:
- Ecology and ethology
- Computer vision and machine learning
- Animal behavior analysis
Background:
- Honey bees are crucial pollinators, and understanding their behavior is vital for ecological studies.
- Tracking individual bees is challenging due to hive complexity, occlusion, and similar bee-to-background appearances.
- Existing methods struggle with the dynamic and dense nature of bee interactions within a hive.
Purpose of the Study:
- To develop and evaluate a robust system for tracking individual honey bees within a beehive environment.
- To address the complexities of high-density object detection, segmentation, and occlusion in bee behavior analysis.
- To provide a reliable method for collecting ecological data through precise bee movement and interaction monitoring.
Main Methods:
- Utilized Mask R-CNN with a ResNet-101 backbone for multi-object detection and segmentation of individual bees.
- Employed a Kalman filter for robust multi-object tracking across image frames, focusing on bee body trajectories.
- Integrated deep learning for detection/segmentation with filtering for tracking, validated using custom observation beehive videos.
Main Results:
- Achieved a mean average precision (mAP) of 0.85 for the Mask R-CNN based multi-object segmentation task.
- The Kalman filter tracking system demonstrated 77.48% MOTA (Multiple Object Tracking Accuracy) and 79.79% MOTSP (Multiple Object Tracking Precision).
- The combined system for tracking and segmentation achieved 77.00% MOTSA (Multiple Object Tracking Segmentation Accuracy) and 80.30% recall.
Conclusions:
- The proposed deep learning and Kalman filter framework effectively tracks individual honey bees in complex hive environments.
- The system successfully overcomes challenges like occlusion and background similarity, providing accurate segmentation and tracking.
- This approach offers a valuable tool for detailed ecological studies of honey bee behavior and social interactions.
