In view of the problems of low detection efficiency， high work intensity， and high error detection rate in the detection methods， e.g.， artificial visual vision and stalk measurement， for the automatic riveting quality detection of launch vehicle capsules， an online detection method for the automatic riveting quality of launch vehicle capsules based on the machine visual system is proposed. With the method， launch vehicle capsules can be detected with high efficiency and high quality while being automatically drilled and riveted. The mapping relationship of a rivet digital image to the physical space is established， and high-quality rivet pier head images are obtained by the angle correction algorithm based on the perspective transformation. The pier head edges and surface images are used to obtain the key size parameters and surface defect information of the pier head， and then the riveting quality is determined. The results of the process test show that the average detection speed of this method is up to 0.92 s/rivet， which can be completed synchronously with the riveting process， and greatly improves the efficiency of riveting quality detection. The detection rate of unqualified rivets is 100%， and the detection accuracy is 99.8%， which significantly reduces the false detection and missed detection rate.