Objective: At present, mechanical screening method is widely used in cigarette capsules size separation, which has some problems, such as low size detection accuracy, unable to separate different color capsules, leakage capsules and so on. Aiming at the quality detection of cigarette capsules, a quality detection system of cigarette capsules is designed and constructed, which integrates the functions of feeding transmission, machine vision image processing and rejection of unqualified products. Methods: In the detection system, the improved inter class variance method was used to extract the single shot image, and the automatic detection of shot quality was realized by gray analysis, unlimited transformation and improved minimum circumscribed circle algorithm. After the automatic image inspection was completed, and it would be rejected by the designed rejecting mechanism for unqualified products. Results: Through repeated experiments and application analysis, the total false detection rate of diameter or shape or worn capsules of unqualified capsules was less than 3%. It was showed that the proposed method could accurately eliminate unqualified capsules, which verified the feasibility and reliability of the system. Conclusion: The design of online detection system for cigarette capsules based on machine vision can complete the automatic online detection of the quality of the capsules, improve the detection speed and accuracy, and have reference value for improving the competitiveness of the market of the explosive capsules.