Daytime star images captured by dedicated near-space star sensors are characterized by short exposures, high noise, and low Signal-to-Noise Ratios (SNRs). Such imaging is also affected by atmospheric turbulence, causing optical phenomena, such as scintillation, distortion, and jitter. This causes difficulty in recording high-precision star images during the daytime. This study proposes an adaptive star point extraction method based on dynamically predicting stars’ positions. First, it predicts the approximate position of stars based on the star catalog, sensor attitude, observation time, and other information, improving the extraction accuracy. Second, it employs a regional SNR sorting method that adaptively selects star images with higher SNRs, suppressing the scintillation effect and enhancing the SNR of star images. Third, depending on the star’s motion trajectory characteristics on the image plane, it utilizes the centroid smoothing method for extraction, thus overcoming the impact of star drift. Field experiments demonstrate that the proposed method can effectively overcome star scintillation, drift, and irregular imaging caused by atmospheric turbulence, achieving a 100% success rate. Moreover, the extraction accuracy improves by more than 80% compared to traditional adaptive methods, attaining a value of 0.05 pixels (0.5"), thereby meeting the requirements of daytime astronomical attitude determination and positioning.