The invention discloses an abnormal behavior sequential detection method based on multi-factor inconsistency. The position, speed and moving direction information is taken into consideration, and the real-time abnormity detection of the target abnormal behavior is realized through an on-line learning and sequential abnormity detection means. The method includes the steps of defining input and output variables, initializing, repeatedly conducting corresponding abnormity detection for each data point of a tested sample and each sample in a training sample sequence, updating the training sample sequence after abnormity detection for each data point of the current tested sample is completed, updating a multi-factor Hausdorff distance matrix, and taking the updated training sample sequence and updated multi-factor Hausdorff distance matrix as new input variables to conduct abnormity detection for the next tested sample. The method is characterized in that the parameters are easy to set, the false alarm rate is controllable, the abnormity detection is highly accurate, the engineering is easy to realize, and the application prospect in the early-warning supervision field is promising.