In-cabin user abnormal behavior detection method applied to ATM

The invention relates to an in-cabin user abnormal behavior detection method applied to ATM. The method includes the following steps: S1: pretreatment and background modeling are performed on an imageand a moving target is extracted and tracked; S2: whether the moving target is a new target or not is judged according to the tracking of the moving target, if the moving target is a new target, a foreground is sent to a personnel classifier subjected to well depth learning training to judge whether the moving target is a person or not, if the foreground is not the human, a count is not changed,otherwise the count is added with 1; if the foreground is an old target, the target is classified by the classifier already, the classifier is not called again; S3: whether a smashing behavior classifier or a tussle behavior classifier is started is judged according to a human number, a moving vector characteristic is extracted and the corresponding classifier is called to obtain a result; S4: andwhether the target leaves