Amirkabir campus dataset: Real-world challenges and scenarios of Visual Inertial Odometry (VIO) for visually impaired people

Visual Inertial Odometry (VIO) is the task of estimating the accurate camera trajectory by using camera and Inertial Measurement Unit (IMU) sensors. There are a wide variety of applications for the VIO, such as augmented reality and indoor navigation. Using VIO methods, a visually impaired person would be able to navigate inside indoor environments and outdoor.... However, dynamic environments such as a crowded corridor are still exceedingly challenging for state-of-the-art VIO algorithms. Existing VIO datasets, e.g., ADVIO, typically fail to effectively exploit challenges. To address this issue and improve visually impaired people’s navigation systems, we introduce the AUTVI dataset. AUT-VI is a novel, generous, and super challenging dataset with 126 diverse sequences in 17 different locations for evaluating VI/VIO algorithms in the context of dynamic objects, robust loop-closure/map-reuse, different lighting conditions, reflections, and sharp camera movements to cover all extreme navigation scenarios. For trajectory evaluation, we provide accurate ground-truth pose from dual GPS and IMU for outdoor and motion capture for indoor. Furthermore, to provide the necessary tools for further development, we made the Android application publicly available, so that fellow researchers can easily record their own variation of the VIO dataset. We also evaluate state-of-the-art VIO and VO methods on our dataset to show the absolute need for this dataset.