The invention discloses a signalized intersection non-motor vehicle illegal cross-street behavior automatic judging method which comprises the following steps: (1) two cameras are set up in a pedestrian crossing and near range of the pedestrian crossing, relative parameters are arranged according to recorded videos; (2) a signalized intersection non-motor vehicle illegal cross-street behavior binary logit prediction model is built according to discrete choice analysis method, standardization is conducted to the binary logit prediction model by adopting a parameter progressive method, the result is marked; and (3) the parameter data are inputted in the binary logit prediction model and therefore signalized intersection non-motor vehicle illegal cross-street behaviors are obtained, when P is larger than 0.5, a non-motor crosses the street illegally; when P is smaller than 0.5, the non-motor does not cross the street illegally. The signalized intersection non-motor vehicle illegal cross-street behavior automatic judging method reduces artificial participation, improves judging speed, and enables the judging to be correct and fair, and meanwhile, provides theoretical foundation for intersection transformation through the factors which obviously influence signalized non-motor vehicle illegal cross-street behavior, and has practical engineering applying value.