Sleeping is one of the most important activities in our daily lives and affect our health. However, very few people could really understand their sleeping habits, which is important to avoid potential sleep-related diseases. Most current studies on sleeping posture studies aim at the monitoring of sleeping postures. However, they are limited to be used in hospitals and need experts to operate these equipment. In this paper, we proposed an automatically sleeping posture estimation system for ordinary people to use in their homes. The customers are only required to wear two sensors, one on chest and the other on wrist during the training process of the sleeping posture monitoring model. We adopted random forest algorithm in the model training algorithm. After this training procedures, users' sleeping postures can be recognized by only wearing one sensor on the wrist. Also, we proposed a data cleaning procedures to process raw sensor data to find the ground truth of sleeping posture. Our experiment results showed that the proposed sleep posture technique can estimate the body posture accurately.