While adaptive splitting the quadtree-structured coding units according to the video content is utmost important in achieving the best encoding performance of the High Efficiency Video Coding (MPEG-H HEVC/H.265), the newest video coding standard, its optimal decision requires much computation. To resolve this complexity problem, we extend the keypoint-based CU depth decision (KCD) method by further providing it with adaptive decision capability to reduce its complexity more. Experimental results show that the adaptive method can save encoding time by 11.6% with sacrificing BD-rate only by 0.68% compared to its non-adaptive scheme.