基于PP-MRF模型的单目车载红外图像三维重建

Three-Dimensional Reconstruction from Monocular Vehicular Infrared Images Based on PP-MRF Model

针对车载红外图像的特点,提出了一种使用超像素分割和面板参数马尔科夫随机场(PPMRF)相结合的单目车载红外图像三维重建方法.该方法首先通过超像素分割得到在纹理和亮度上相近的一系列小的区域,即超像素,然后训练PP-MRF模型,使它能对待测试图像的各个超像素进行面板参数的分析和深度估计.通过实验证明了该方法能够有效地对单目车载红外图像做出深度估计及三维重建.

A three-dimensional reconstruction method of monocular vehicular infrared image, which combines super pixels segmentation and the plane parameter-Markov Random Field （ PP-MRF ） model, is proposed based on the characteristics of the vehicular infrared image. Firstly, the image is segmented into a series of small areas where the texture and brightness are similar, i.e. super-pixels. Then the PP- MRF model is trained, which can analyze the plane parameters and estimate the depth of each super- pixel of the testing image. The experimental results show that the proposed method can estimate the depth value of monocular vehicular infrared images and rebuild the 3D scene properly.