CMDPC_OBB: A Large-Scale Image Dataset for Coal Mine Drill Pipe Counting based on Oriented Bounding Box.

Area of Science:

• Mining Engineering
• Computer Vision
• Data Science

Background:

• Limited availability of public datasets for underground coal mine monitoring hinders accurate drill pipe recognition.
• Complex drilling environments present significant challenges for object detection and counting tasks.

Purpose of the Study:

• To construct a large-scale, publicly available image dataset (CMDPC_OBB) for coal mine drill pipe counting.
• To address the scarcity of data and improve recognition accuracy in challenging underground mining conditions.
• To facilitate multi-pose and multi-view drill pipe detection and counting.

Main Methods:

• Developed the CMDPC_OBB dataset comprising two sub-datasets: MOD_2D (114,869 images with oriented bounding boxes) and SOC_3D (4,023 images with 3D reconstruction for enhanced multi-view representation).
• Employed a 15-frame interval sampling strategy for MOD_2D and single-image 3D reconstruction for SOC_3D.
• Evaluated nine object detection and oriented detection models on the CMDPC_OBB dataset.

Main Results:

• The CMDPC_OBB dataset provides extensive data coverage for complex underground mining scenarios.
• The highest mean Average Precision (mAP) achieved was 89.1% on the CMDPC_OBB dataset.
• The dataset demonstrates significant effectiveness and benchmarking value for drill pipe detection and counting.

Conclusions:

• The CMDPC_OBB dataset effectively addresses the need for robust data in underground coal mine monitoring.
• The developed dataset and evaluation results pave the way for improved automated drill pipe counting systems.
• This resource is crucial for advancing research and development in mining safety and operational efficiency.