The union method of Isomap and SVR is proposed for exploring modon trajectories of moving objects．Isomap is one of the famous NDR methods，is rapidly in implementation，docsn’t require extra preprocessing for input data,can give the intra dimensionality of high dimensional nonhnear input data，and has the capability of providing low dimensional representation which is useful for finding motion trajectories．But the modon trajectories from Isomap output are only analogous motion trajectories，not real motion trajectories．So a nonlinear mapping from Isomap output to real trajectories must be established to gain correct motion trajectories．SVM is one of the excellent machine learning methods with predominant ability of small samples learning．When SVM is used for regression or function estimation，it is called SVR．SVR is adopted for calibrating Isomap output．Calibration error vector and calibration error are defined as the measurement of adjusting quality．The comparison between SVR and other methods is also studied．Experiments resuls show,proposed method is effective and correct．
展开▼
