When applied inside Earth's atmosphere, the star tracker is sensitive to sky background produced by atmospheric scattering and stray light. The shot noise induced by the strong background reduces star detection capability and even makes it completely out of operation. To improve the star detection capability, an attitude-correlated frames adding (ACFA) approach is proposed in this paper. Firstly, the attitude changes of the star tracker are measured by three gyroscope units (GUs). Then the mathematical relationship between the image coordinates at different time and the attitude changes of the star tracker is constructed (namely attitude-correlated transformation, ACT). Using the ACT, the image regions in different frames that correspond to the same star can be extracted and added to the current frame. After attitude-correlated frames adding, the intensity of the star signal increases by n times, while the shot noise increases by n~n/2 times due to its stochastic characteristic. Consequently, the signal-to-noise ratio (SNR) of the star image enhances by a factor of n~2n. Simulations and experimental results indicate that the proposed method can effectively improve the star detection ability. Hence, there are more dim stars detected and used for attitude determination. In addition, the star centroiding error induced by the background noise can also be reduced.