To address the issue of increased parameter and computational complexity leading to decreased efficiency in improving prediction accuracy for human pose estimation models, a lightweight human pose estimation network called MC2Pose was designed based on the YOLOPose model. The approach involved rebuilding the backbone network using the lightweight MobileNetV3 network to maintain feature richness while speeding up feature extraction. Additionally, the UPSample module was replaced with the CARAFE module to perform upsampling.We also introduce the CBAM module to avoid the problem of reduced model accuracy caused by lightweight optimization. Experimental results demonstrated that the YOLOpose model, after the aforementioned lightweight improvements, reduced its model size to 134.9MB, a decrease of approximately 16.2%, and reduced GFLOPS to 87.3, a decrease of approximately 14.5%. The model's computational complexity significantly decreased, and the addition of the CBAM attention mechanism had minimal impact on model accuracy. Overall, these improvements ensure both accurate recognition and lightweight detection algorithm implementation in the improved model.