# Fast Three-Dimensional Profilometry with Large Depth of Field

## TL;DR

This paper introduces a new method for fast 3D imaging that uses a neural network to improve speed and depth of field.

## Contribution

The novel contribution is a time-domain Gaussian fitting method combined with a neural network for rapid 3D profilometry.

## Key findings

- The proposed method extends the system's depth of field by five times.
- Data acquisition and computing times are reduced to under 35 ms.
- The method works well on complex surfaces without deformation.

## Abstract

By applying a high projection rate, the binary defocusing technique can dramatically increase 3D imaging speed. However, existing methods are sensitive to the varied defocusing degree, and have limited depth of field (DoF). To this end, a time–domain Gaussian fitting method is proposed in this paper. The concept of a time–domain Gaussian curve is firstly put forward, and the procedure of determining projector coordinates with a time–domain Gaussian curve is illustrated in detail. The neural network technique is applied to rapidly compute peak positions of time-domain Gaussian curves. Relying on the computing power of the neural network, the proposed method can reduce the computing time greatly. The binary defocusing technique can be combined with the neural network, and fast 3D profilometry with a large depth of field is achieved. Moreover, because the time–domain Gaussian curve is extracted from individual image pixel, it will not deform according to a complex surface, so the proposed method is also suitable for measuring a complex surface. It is demonstrated by the experiment results that our proposed method can extends the system DoF by five times, and both the data acquisition time and computing time can be reduced to less than 35 ms.