In order to realize the real-time image stabilization of the UAV aerial image and solve the problem of slow speed in feature detection and the divergence in motion filtering during the process of image stabilization, a real-time image stabilization algorithm is proposed combining the improved AGAST (Adaptive and Generic corner detection based on the Accelerated Segment Test) algorithm with the adaptive Kalman filter. For the real-time aerial video sequences, the image stabilization processing is based on the frame prior to the current frame. The improved AGAST feature detection algorithm rapidly extracts the features of AGAST corners on the basis of scale space, which are described by binary descriptors, and then the feature points are matched by using Hamming distance. For the matched sets of point pairs, the RANSAC principle is used to eliminate the mismatched points and then the motion vector is calculated. Finally, the adaptive Kalman filter is used to extract the active scanning component of the UAV in the motion vector, which performs motion compensation to obtain the result of image stabilization. The standard test atlases and the UAV aerial video collected by researchers serve as the experimental data.The experimental results show that the proposed algorithm is effective and fast in the processing of the continuous video sequenceand can meet the needs of real-time image stabilization.decision tree; scale space; motion filtering; Kalman filtering