In this paper, we present the implementation and comparison of various human pose estimation algorithms on power-constrained devices with limited GPU capabilities and resources, specifically for smart eyewear applications. First, we reviewed existing pose estimation algorithms and their deployment on edge devices, with a particular focus on the Jetson Orin Nano due to its portability, low power consumption, and robust computing capabilities. Our initial tests with RTMPose on the Jetson Orin Nano revealed it was unsuitable for real-time processing. To address this issue, we employed MMDeploy to optimize RTMPose, enabling real-time performance and the ability to process more than six persons per frame. Despite the RTMPose’s reliance on the complex MMCV library, it significantly outperformed YoloPose v5 and YoloPose v8 in terms of both inference speed and average precision. This optimized version of RTMPose stands out as the best candidate for deployment in smart eyewear applications for human pose estimation.