The invention relates to video processing, mode identification and the like. In order to provide a video abnormal event detection method by which a multi-motion parameter variation rule of the same event in a plurality of visual angles of a cross camera can be represented and of which the accuracy is high, the technical scheme of the invention is that: a method for detecting a monitored video abnormal event based on trace analysis comprises the following steps of: 1, classifying abnormal events based on the trace analysis; and 2, modeling the abnormal events based on a dynamic Bayesian network, wherein the modeling is finished by the following three steps of: a, constructing the dynamic Bayesian network which consists of three levels from bottom to top, namely a characteristic layer, an element layer and an event layer; b, learning the dynamic Bayesian network in the following three situations: firstly, the labels and the number of elements are known; secondly, the number of the elements is known and the labels are unknown; and thirdly, the number of the elements and the labels are unknown; and c, selecting a characteristic. The method is mainly applied to video processing, mode identification and the like.