DynaVIG: Monocular Vision/INS/GNSS Integrated Ego and Object Localization in Dynamic Scenes

Existing ego and object localization systems, predominantly reliant on stereo or stereo-inertial methods, encounter large computation demands and are prone to error drift. These systems often assume a constant camera height and constant object velocity model, presupposing a planar and straight-line ground. We propose DynaVIG, a unified framework of ego and object localization system based on the integration of Monocular Vision, Inertial Navigation System (INS), and Global Navigation Satellite System (GNSS), aiming to provide accurate estimation of the ego and object poses with low computation and cost. A prior height model and an accurate dynamics model are constructed to forego the conventional ground assumptions. Experiments on the KITTI dataset demonstrate that our framework can achieve accurate ego and object localization with high efficiency compared to state-of-the-art methods.