EMA-VIO: Deep Visual–Inertial Odometry With External Memory Attention

Accurate and robust localization is a fundamental need for mobile agents. Visual–inertial odometry (VIO) algorithms exploit the information from camera inertial sensors to estimate position translation. Recent deep-learning-based VIO models attract attention as they provide pose in data-driven way, without of designing hand-crafted algorithms. Existing learning-based rely on recurrent fuse multimodal data process sensor signals, which are hard train not efficient enough. We propose novel framework with external memory that effectively efficiently combines visual features state estimation. Our proposed model able accurately robustly, even challenging scenarios, example, overcast days water-filled ground, difficult traditional extract features. Experiments validate it outperforms both baselines different scenes.