Year: 2025
Type: article
Abstract: The complex and unique operational environment of underground mines presents significant challenges for personnel target detection and tracking. Workers typically wear standardized work uniforms, resulting in highly similar visual appearances. In addition, insufficient illumination, motion blur, and monotonous gray backgrounds significantly reduce the accuracy of visual recognition. In typical underground mining scenarios, high personnel density and frequent occlusions further increase the difficulty of tracking algorithms. In deep learning model training, the quality of datasets often plays a decisive role determining model performance. However, there is currently a lack of publicly available and standardized image datasets for personnel tracking in mining environments, both domestically and internationally. Most existing studies rely on small-scale, non-standardized private datasets, which are insufficient for comprehensive algorithm benchmarking and large-scale application. To address this gap, this study utilizes intrinsically safe law enforcement recorders designed for coal mines to capture real-world occlusion scenarios in underground mining environments. Based on the collected data, a large-scale, standardized, and high-quality occlusion-aware personnel target tracking dataset for mines is constructed. The dataset comprises a total of 14,740 images with an overall size of 38.1 GB, covering three typical occlusion-prone scenarios in mines: the exit gate at the shaft entrance, the well chamber at the bottom of the shaft, and the downhole working face. The establishment of this dataset effectively alleviates the scarcity of occluded personnel tracking resources in mining environments and holds high practical value. It provides essential data support for related research and facilitates the development and evaluation of advanced tracking algorithms in challenging real-world mining scenarios.
Source: China Scientific Data
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7118118073