Abstract: Since the traditional method could not effectively monitor the screw locking device of urban rail transit door lock because of long-term reciprocating impact, a new monitoring method based on machine vision measurement is proposed. Firstly, the screw locking device is marked with a red dot and a red line, which will be detected respectively to calculate the angle of rotation deviation. Secondly, the images of screw locking device collected by camera are pre-treated by DCT (discrete cosine transformation), the threshold segmentation is used to gain binary images, then to calculate the angle information by capturing the featured marks. Finally, the statistical data analysis method is used to judge and predict the train door lock safety. Experimental results show that the dot marks are detected easier than line marks in collecting more accurate angle information.