Abstract: In consideration of the visual data and the data from the inertial measurement unit (IMU) having the different characteristics, the 6D pose of the visual sensor is recovered from the captured video by using the vision-based 3D reconstruction techniques, the obtained pose information and inertial measurements is fused via the spatial-temporal alignment, and based on the determined scale of the visual 3D reconstruction the non-contact measurement of a scene is performed. The proposed approach achieves a relative measurement error of about 3% in the non-contact object size estimation experiment. As it does not require the additional calibration object or any special motion capture devices, the proposed method is flexible and is applicable in many potential areas.

Key words: visual-inertial sensing data, 3D reconstruction, spatial-temporal alignment, non-contact object size acquisition