An IMU-Assisted Robust Alignment Algorithm Under Dynamic Foreground Interference

Accurate alignment of the background region can faithfully reflect the true motion of the camera， playing a critical role in applications including video stabilization and image mosaicking. However, dynamic foreground interference—particularly subtle disturbances and large-scale motion—remains one of the major challenges to achieving stable and precise alignment. To address this, we propose an Inertial Measurement Unit (IMU)-assisted robust image alignment algorithm designed to effectively suppress dynamic foreground interference and outlier points, thereby enhancing alignment accuracy. The method first estimates the camera’s three-axis rotation using IMU data and then constructs a motion analysis model based on inter-frame residual amplification derived from the observed displacement of feature points and their rotation-compensated offsets. Residual velocity cues are further incorporated, and a clustering-based strategy is employed to identify and suppress subtle disturbances and abnormal motions. To enhance the reliability of background point identification, the spatial compactness of feature points is also taken into account. Once the background points are determined, the translational component of the camera motion is estimated accordingly, and an error-feedback mechanism is introduced to iteratively refine the rotation parameters, thereby improving the overall transformation accuracy. Experimental results demonstrate that the proposed method delivers superior alignment performance and robust background extraction under challenging conditions with dynamic foreground interference.