Can smart devices assist in geometric model building?

Abstract

The creation of precise three dimensional geometric models produced from visual information from low cost smart devices has many potential applications. There is a need to establish the capability of such devices, to determine if they are suitable. We explain how a typical smart device could be used in the creation of precise 3D geometric models from visual information, and establish a benchmark for evaluating such devices via a simulator which allows for parametric exploration of device capabilities. Various vision based algorithms (e.g. SLAM) could benefit from auxiliary sensor inputs, e.g. to solve the problem of determining the scale of the object being sensed. We discuss how the ancillary sensors in smart devices can assist in 3D mensuration to underpin 3D scene generation and 3D model building. We describe a number of tests performed on the inertial sensors integrated into consumer smart devices to ascertain their performance characteristics. These findings are then incorporated in our simulations. We also explore the extent to which future improvements to technologies are likely to be beneficial. Our simulator allows modelling of a 3D scene, various sensors and the motion trajectory of the device whilst capturing data. We assess individual sensor outputs, and the quality of the fused sensor outputs in the context of 3D model building requirements.