MOVRO: Loosely Coupled EKF-Based Monocular Visual Radar Odometry

Reliable and accurate ego-motion estimation constitute an integral part of any autonomous system. To achieve this, it is essential to integrate sensors of complementary characteristics while also minimizing costs. The most often used ego-motion estimation setup usually consists of a monocular camera and an inertial measurement unit (IMU) which can often drift due to inaccuracy of modest performance IMUs. Inlieu of enhancing such setups with a higher-performing IMU, in this paper we propose to fuse the monocular camera with a radar sensor. Specifically, we combine measurements from a 3D millimeter-wave frequency modulated continuous wave radar and a monocular camera within an error-state extended Kalman filter framework. The filter was implemented using loose coupling that allows for an independent and scalable development of sensor models. We evaluated the proposed method against state-of-the-art radar inertial, visual inertial, and radar visual inertial odometries. The results show that the proposed method is comparable with other multi-domain solutions, especially in outdoor sequences.