<p>Human pose estimation is a fundamental task in computer vision, aiming to accurately localize body keypoints, and reconstruct skeletal structures. However, due to significant pose variations and frequent occlusions, existing networks still struggle to extract robust pose representations. Furthermore, current multi-resolution feature fusion methods often suffer from semantic misalignment across different scales. To address these challenges, we propose an Enhanced Context-aware Cross-Resolution Fusion Network (ECCFNet). First, we design a Multi-Scale Context-Aware Block (MSC-Block) to capture diverse pose features more effectively. This block enhances the modeling capability for complex poses while maintaining a low parameter count. Second, to resolve feature inconsistencies during multi-resolution fusion, we propose a Dynamic Upsampling Cross-Resolution Feature Fusion Module (CRFFM), which improves keypoint localization accuracy via adaptive aggregation. Finally, to mitigate spatial information loss during feature propagation, we introduce an Efficient Multi-Scale Attention (EMA) module. This module integrates information from distinct spatial locations through cross-spatial learning, enhancing both local detail representation and global context modeling. Extensive experiments on the COCO 2017 and MPII datasets demonstrate that ECCFNet reduces parameters by 26% and computational cost by 51% compared to HRNet, while achieving a 1.5-point AP improvement. These results highlight ECCFNet’s practical advantages, specifically its suitability for resource-constrained edge deployment and its superior robustness in handling heavily occluded or low-resolution inputs. Our method outperforms both the baseline and state-of-the-art approaches, validating its effectiveness and superiority in human pose estimation tasks.</p>