The invention discloses a worker violation behavior monitoring method, and the method comprises the steps: designing a multi-scale prediction target detection model, carrying out the prediction of a violation target at four scales, and obtaining the category information and position information of the violation target. According to the method, the deep-layer features and the shallow-layer featuresare fused by using a double up-sampling rate, and the target is detected from different scales, so that the method has relatively high accuracy. Besides, the method provided by the invention can takealarm measures for workers having violation behaviors according to the detection result; the workers who detect the violation behaviors are tracked; according to the method, whether the workers having violation behaviors in two adjacent frames of video images are the same person or not is judged, the situation that the violation behaviors of the same violation worker are recorded multiple times due to the fact that only a single frame of image is considered is avoided by considering the time domain information, violation information processing of managers is facilitated, and the method is a more accurate monitoring method.