In order to improve the dynamic measurement range and measurement accuracy of the Hartmann-Shack Wavefront Sensor (HSWS) in an adaptive optics system, the centroid detection for a wavefront spot pattern collected by a CCD is researched. The principle of HSWS is introduced and an adaptive centroid detection method by dynamic positioning spot region, dynamic segmenting threshold level and locking optimal detection window is presented according to the structure features of HSWS with 127 units and the distortion characteristics of actual human eye spot pattern.The effects of threshold level selection on human eye spot centriod detection are discussed and the centroid detection accuracy and noise robustness of the method are analyzed through simulation and experiments. Experimental results indicate that selecting the threshold level to be (50±3)% of the total grade amount is more appropriate and the centroid detection error can be decreased by 60% or more as compared with that of other detection methods. By using adaptive centroid detection method in the HSWS of an adaptive optical retina imaging system, human eye aberration can be reduced from 0.728λ to 0.081λ in RMS(λ=785 nm) through a closed loop correction.Moreover, the system can reach the diffraction limit basically and can obtain a retina image. These results show that the adaptive centroid detection method can select different experimental parameters aimed at the characteristics of spots, which makes up the limitations of the general methods.It meets the feasibility and practicability of human eye aberration measurement.