The electrocardiogram (ECG) contains many significant information of feature values, which can be used to reflect human physiological health, and is an important indicator for diagnosing cardiovascular diseases. The wearable ECG monitoring equipment provides patients with long-term cardiogram monitoring, but the acquisition signal is susceptible to artifact noise contamination. Reducing motion noise while ECG signals processing will help accurately analyze the patient's cardiovascular signal and make correct warning and judgment on patients. This paper mainly analyzes how to enhance the ECG collected under long-term monitoring, and try to propose an adaptive ECG enhancement method. The method is composed of the feature extraction of ECGs, adaptive detection of human motion state and a modified Wiener filter based on Bayesian estimation. The method is evaluated on MITDB and CPSC2019 database, as well the synchronous ECGs and three-axis acceleration data in the real world. The heart rate performance index is designed, and it is found that the estimation accuracy of heart rate can be improved by 24.5% for the enhanced ECG signals from a strong noise environment. It is proved that the method can achieve good performance of ECG signal enhancement in different body motion states.