<p>Landslide Early warning systems has been widely used to avoid potential disaster. In this paper, a fast monitoring and real time precursor predication method is proposed to build the early warning systems for specific landslide. The fast monitoring network in this system uses ad-hoc technology to build rapid site monitoring network consist of Beidou terminals and fracture monitors. The real time precursor predication method based on the KF-FFT-SVM model is conducted to fulfil precursor early warning of in short time. The KF-FFT-SVM model working in this system is established through the analysis of the precursor slide character in deformation data got by the Beidou terminals. The deformation data is considered as the mechanical vibration of specific landslide and the KF-FFT-SVM model is trained to predicate the occurrence of landslide by the real time deformation data. This system not only improves the robustness of site monitoring, but also provides an effective early warning method for specific landslide. It is applied in Baige landslide monitoring and results showed that KF-FFT-SVM early warning model can predication the occurrence of landslide with high accuracy. It will make the early warning work for specific landslide more effective and costless, although numerous continuous monitored precursor slide deformation data are needed to trained the model well.</p>