Due to the limited memory and computing resources in low performance edge computing, traditional large models are difficult to meet the high demand for latency in computational tasks. This paper builds upon the Pose ResNet18 Body model for human pose estimation, and analyses the hardware limitations of the low performance embedded device Jetson Orin Nano 4G, including limited memory and processing capability, as well as strict requirements for energy consumption and power. Then two model compression methods are proposed to address these limitations: compression based on INT8 quantization and compression based on weight sparsity pruning. Finally, the compressed models are validated on the Jetson Orin Nano 4G platform, and experimental results demonstrate significant advantages in inference time, storage space, power consumption, while maintaining the required prediction accuracy.