Aiming at the low efficiency and poor stability of crankshaft roundness measurement,a measurement method based on machine vision was proposed.A vision platform was built to collect images;image preprocessing and sub-pixel method were used to improve the edge accuracy,and the circle was reconstructed by 3D reconstruction.Then,the visual evaluation scheme of crankshaft roundness error was proposed,and the measurement accuracy was improved by tilt correction and rotation error compensation.Finally,the contrast experiment of crankshaft roundness error measurement was carried out.The results show that the average error of visual evaluation results is 6 μm.The roundness error measurement of the crankshaft can be realized and more stable.