Fusion of Vision and Inertial Data for Motion and Structure Estimation

This paper presents a method to fuse measurements from a rigid sensor rig with a stereo vision system and a set of 6 DOF inertial sensors for egomotion estimation and external structure estimation. No assumptions about the sampling rate of the two sensors are made. The basic idea is a common state vector and a common dynamic description which is stored together with the time instant of the estimation. Every time one of the sensor sends new data, the corresponding filter equation is updated and a new estimation is generated. In this paper the filter equations for an extended Kalman filter are derived together with considerations of the tuning. Simulations with real sensor data show the successful implementation of this concept.