@Article{ee.2026.084197,
AUTHOR = {Lingwen Meng, Jintong Ma, Siqi Guo, Guanghui Xi, Siwu Yu},
TITLE = {Zoom-Adaptive Personnel Tracking for Intelligent Energy Infrastructure and Reliable Power Grid Operation},
JOURNAL = {Energy Engineering},
VOLUME = {},
YEAR = {},
NUMBER = {},
PAGES = {{pages}},
URL = {http://www.techscience.com/energy/online/detail/27989},
ISSN = {1546-0118},
ABSTRACT = {Reliable operation of intelligent energy infrastructure increasingly depends on continuous, identity-consistent awareness of personnel working near critical grid assets. Zoom-adaptive cameras enable flexible switching between site-wide situational awareness and close-range inspection, but focal-length changes alter person scale, resolution, and apparent motion, causing conventional trackers to fragment trajectories and confuse identities. This study develops a zoom-adaptive global identity association framework for cross-scale tracking of power-grid operation personnel. Focal length is modeled as an explicit physical state throughout geometric normalization, feature learning, and trajectory inference. The framework integrates focal-length-normalized coordinates, multi-scale identity embeddings, grid-specific safety attributes such as helmet and workwear cues, and graph-based association with temporal smoothing. Experiments use a power-grid surveillance dataset comprising more than 180,000 frames, 1200 identities, focal lengths of 5–80 mm, and representative occlusion, illumination, and crowding conditions. Against DeepSORT, ByteTrack, OC-SORT, BoT-SORT, TransTrack, and a ReID-only association baseline, the proposed method increases the identity F1 score (IDF1) by approximately 8%–13% and reduces identity switches and trajectory fragmentation by more than 25% under substantial zoom variation. Ablation and robustness analyses identify explicit zoom modeling as the principal source of identity stability, with motion consistency, semantic attributes, and temporal smoothing providing complementary gains. Edge-deployment analysis further shows how crop-level processing, quantization, local graph inference, and metadata-level communication can limit latency, bandwidth, and computing demand. By sustaining reliable personnel trajectories across wide-area observation and detailed inspection, the framework provides an AI-enabled perception layer for safety-zone enforcement, maintenance accountability, and rapid incident response. These capabilities strengthen reliable and resilient power-grid operation while advancing the resource-efficient digitalization of next-generation energy infrastructure.},
DOI = {10.32604/ee.2026.084197}
}