Visual-inertial odometry (VIO) is a technique to estimate the motion of a vehicle platform by fusing camera and inertial sensor data. It operates effectively in GPS-denied environments such as indoors and is widely utilized in applications like autonomous navigation of unmanned aerial vehicles (UAVs) due to its real-time performance and high localization accuracy. However, since VIO relies on textures in the environment or features extracted from image frames, localization may easily fail if the number of feature points in the image is insufficient or the U AV faces a low-texture environment. To address these issues, we propose an active VIO algorithm by planning heading angles autonomously. This algorithm improves VIO accuracy and maintains robust localization even in an unknown environment by employing heading planning to acquire more feature points in the subsequent image frames. To achieve this, we first divide an image frame into several sections and count the number of feature points in each section. Next, we determine the desired heading angle based on the feature-occupied ratio of each section. The proposed approach is validated in various cases in a simulation environment that mimics an indoor warehouse.