基于深度学习的光场成像三维测量方法研究

Three-Dimensional Measurement Method of Light Field Imaging Based on Deep Learning

为了解决光场相机应用于三维测量时,在弱纹理区域和精细结构区域难以获得准确视差估计结果问题,提出了基于深度学习技术对光场深度估计进行建模,并建立了光场视差与真实深度之间的转换关系。将所提方法应用于多种复杂场景中,实验结果均表明:该方法可以准确获取弱纹理区域和精细结构区域的视差信息,较好地复原场景的三维结构,视差估计处理时间压缩到1 s量级,相比传统的基于代价优化的方法,降低了1~2个数量级。

To estimate the accurate disparity in weak texture region and fine structure region when the light field camera is used for three-dimensional measurement, a model of the light field depth estimation based on deep learning technology is proposed. Moreover, the relationship between the disparity and corresponding depth is also established. The proposed method is applied to a variety of complex scenes, and the experimental results show that the proposed method can accurately estimate the disparity information in the weak texture region and fine structure region, and leading to a good reconstruction of three-dimensional structure. The processing time of the proposed method is compressed to the order of 1 s, which is 1 to 2 orders of magnitude lower than the traditional methods based on cost optimization.