Due to the characteristics of meshing contact and inner package of gear pair, the detection of fatigue strength is greatly limited, especially, the fatigue failure state monitoring such as fatigue crack and gear breakage is usually conducted by stopping at intervals and inspecting after disassembly. The whole process is time-consuming and laborious, the observation errors are caused from manual subjective judgment and experience differences, moreover, the micro-fatigue characteristics of gear in early fatigue stage cannot be perceived. Therefore, a non-contact detection method for gear fatigue strength test was proposed based on binocular vision technology and digital image correlation theory. A reducer test platform with binocular cameras was set up through the transmission means of dual motors to drag, high-quality speckles were prefabricated on gears for fatigue test, the fatigue characteristic images of whole life cycle were collected by using binocular vision method. The target areas on the gear teeth were searched and tracked by using the methods of image matching algorithm, so that the displacement or strain of data points in the area were analyzed and calculated. The results show that by using the proposed method, the gear fatigue test process can be monitored online and the fatigue failure characteristics can be accurately identified such as early micro-cracks without disassembly, which has features of high detection accuracy and strong anti-interference ability that benefit to improving the test efficiency and reducing costs.