Photorealistic Rendering and Modification of Motion Blur for the Simulation of Video Data for Structural Dynamics

Abstract

In recent years imager-based approaches to structural dynamics measurements have gained increasing interest. Imager-based approaches to measurement have a number of attractive properties including being able to monitor large areas at a relatively high spatial resolution with a relatively small number of imagers. The stand-off monitoring capabilities of imagers are also highly attractive for many structural dynamics and structural health monitoring applications, particularly those in dangerous, inaccessible, extreme, and high temperature environments. For their advantages though, a number of challenges must be addressed when using imager-based techniques for measuring the deformation and motion of structures. There are issues of lighting conditions that can potentially vary during the measurement period, due to movement of the sun or lights in a facility being turned on and off. In addition, video measurement of dynamic structures also requires making a number of choices associated with the measurement setup including location of the imagers, focal lengths of lenses, numberof- pixels, lens characteristics, framerates, shutter speeds, ISO, aperture settings, planes of focus, and depth-of-field. There can potentially be effects such as specular reflections or shadows which might complicate downstream data processing. The large number of parameters associated with imager measurements make them very flexible, but it also can mean that setting up imager measurements can take substantial time even under controlled laboratory conditions. The complication associated with imager measurements suggests that highfidelity, photorealistic rendering tools that are capable of capturing the interactions between structural dynamics, light transport, and the measurement process at the imaging plane are needed. To date, substantial work has been done by the computer graphics community to develop photorealistic rendering tools which are becoming increasingly accessible. However, in structural dynamics we often care about sub-pixel motion and it is not understood whether current techniques for modeling effects such as motion blur have sufficient fidelity for structural dynamics. Furthermore, photorealistic video renders can take a large amount of time to complete, and given current techniques do not allow for simple changes such as frame-rate/shutter speed in post processing without redoing the entire render. In this work we present an approach based on digital coded exposures for forming frames of dynamic scenes, that allows for the framerate and shutter speed to be changed in post-processing. The proposed approach is inspired by the physical operation of conventional cameras. In addition, the digital coded exposure proposed approach suggests theoretical alternatives to data capture that could potentially be used to control motion blur properties on a per-pixel basis.