The 3D keypoint selection method has a significant impact on the performance of keypoint-based 6D pose estimation. State-of-the-art selection methods do not perform well in textureless regions and uneven texture distribution, therefore, we propose the GAST-FPS 3D keypoint selection method. Based on the 2D feature points obtained from the Superpoint detector and their confidence, we devise the Global Sliding Adaptive Threshold (GAST) to select the highly salient 2D keypoints on the object surface including the textureless area. Then, the selected 2D keypoints are converted to 3D space and fed into the FPS algorithm to generate uniformly distributed 3D keypoints. The experimental results on the YCB-Video dataset show that our approach outperforms relevant methods and improves the overall accuracy and robustness of keypoint-based 6D pose estimation.