Vision-Inertial Fusion Pose Estimation for UAVs in Complex Environments: A Review and Comparative Analysis

Drone attitude estimation is very sensitive to blocked satellite navigation signals and sensor errors in complex environments. So, methods for autonomous attitude estimation using onboard sensors have gotten a lot of attention. Visual sensors and inertial measurement units complement each other well in the information they provide. Because of this, visual-inertial fusion has become a key research area for drone attitude estimation. This paper reviews and compares UAV visual-inertial fusion attitude estimation methods for complex environments. It organizes typical fusion frameworks based on filtering and optimization. It focuses on the modeling ideas, IMU pre-integration methods, and the features of tightly coupled and loosely coupled fusion strategies in optimization-based methods. Then, under hard visual conditions like low texture, changing light, and dynamic scenes, it looks at how stable and practical different fusion strategies are. The paper lists the main challenges and future directions for current visual-inertial fusion methods.