The work aims to design a single-cut system based on machine vision for rounded bags to solve the problems of low efficiency, large amount of waste and fast mechanical wear of the double-cut method often used in production of rounded bags by bag making machines. A visual hardware platform was built. Based on the OpenCV visual library, visual software was written in the visual studio development environment. The camera was set in the flying shooting mode, and the output signal was divided to the PLC through the first traction servo encoder. The PLC conducted hard trigger on the camera through the I/O port by virtue of a high-speed counter. A calibration board was used to calibrate the camera to get the pixel equivalent value. After filtering, binarization, edge detection and other processing, the feature of the rounded corner was extracted to make the template. By traversing the image, key points were added on the center line of the template corners, the template center coordinates can be directly obtained when the template matching was completed. The difference between the center coordinates of the corners and the baseline coordinates was sent to the PLC which controlled the tool-shifting servo to realize single-cutter precise cutting. This method can realize fast and accurate positioning of fillet center in pulling the material, and the current bag-making cycle was used to move the knife to cut the current bag. The actual cutting error was within ±0.1 mm, which met the current production requirement on minimum fillet radius 0.125 round corner bag. The vision system adopting the flying mode has high image processing speed, can match the high-speed bag-making requirements, effectively solve the problems that occur in the traditional production of round-cornered bags, meet the needs of industrial production, and have great practical application value.