Related Experiment Video
Updated: Apr 8, 2026
08:27
Author Spotlight: Efficient Image Recognition Using Directional Gradient Histogram Technique and Support Vector Machines
Published on: January 5, 2024
1.8K
Improved Chunked Image Annotation Method for Electric Bikes in Complex Elevator Scenarios Based on Local Features
Jixing Yan1, Allam Maalla2, Zehong Hong1
1School of Engineering, Guangzhou College of Technology and Business.
Journal of Visualized Experiments : Jove
|April 6, 2026
Summary
A new chunked annotation method improves electric bike (EBike) detection in elevators by focusing on local features. This approach enhances object detection accuracy and interpretability, crucial for safety-critical AI applications.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Machine Learning
Background:
- Electric bikes (EBikes) in confined spaces like elevators pose safety risks.
- Automated object detection struggles with occluded EBikes in complex scenes.
- Traditional holistic annotation methods are insufficient for partially occluded objects.
Purpose of the Study:
- To develop a novel, interpretable annotation strategy for EBike detection.
- To improve the robustness of object detection models under occlusion.
- To create a specialized dataset for EBike detection in elevator environments.
Main Methods:
- Proposed a chunked annotation method decomposing EBikes into key regions for local feature learning.
- Developed the EBike-DET dataset with chunked annotations and simulated environmental conditions.
- Evaluated detection performance using YOLOv5, YOLOv10, and SSD models.
Main Results:
- The chunked annotation method significantly improved precision, recall, F1 score, and mAP for YOLOv5 on the EBike-DET dataset.
- EBike-DET dataset demonstrated superior stability and robustness against occlusion compared to public datasets.
- The approach enhances explainable AI (XAI) by providing structural interpretability.
Conclusions:
- Chunked annotation is effective for improving EBike detection accuracy and robustness in occluded scenarios.
- The EBike-DET dataset provides a valuable resource for research in object detection within challenging environments.
- This work contributes to more transparent and interpretable AI solutions for real-world safety monitoring.
