Efficient Human Pose Estimation Using a Lightweight Network with ECA Attention
Ramakrishnan Raman, Pooja Sapra, K Nisha, V. Prabakaran · 2025
Human pose estimation is a crucial task in computer vision, with applications in healthcare, sports analytics, and human-computer interaction. Despite significant advancements, existing methods often demand substantial computational resources, limiting their deployment on resource-constrained devices. This paper introduces an efficient and lightweight network architecture for human pose estimation, designed to address the challenges of accuracy and computational overhead. The proposed model leverages a streamlined backbone network, MobileNetV2, known for its efficiency, combined with the Efficient Channel Attention (ECA) mechanism to enhance the representation of spatial and channel-wise features. By embedding ECA attention, the network adaptively focuses on significant features, enabling precise keypoint localization while maintaining computational efficiency. Experimental evaluations conducted on standard datasets, including the COCO Keypoint Dataset and MPII Human Pose Dataset, demonstrate that the proposed model achieves a competitive balance between accuracy and speed. The architecture attains a high average precision (AP) while significantly reducing the number of parameters compared to state-of-the-art methods like HRNet and OpenPose. Furthermore, the lightweight design enables real-time performance on edge devices, making it highly suitable for applications requiring quick response times. This work contributes to bridging the gap between high-performance pose estimation and practical deployment in real-world scenarios, ensuring broader accessibility and usability.