激光与视觉融合的透视平面检测与深度预测研究

Research on Glass Plane Detection and Depth Prediction Based on the Fusion of Laser and Camera

摘要:

针对单一的激光传感器或视觉传感器无法检测到透视三维平面的问题，提出一种基于激光传感器与视觉传感器融合的透视平面检测与深度预测算法。首先采用GDNet，在二维彩色图像中对透视平面进行图像分割；其次应用单一图像反射去除算法，在分割得到的透视平面区域分离背景信息，并使用MegaDepth算法进行深度预测，得到相对深度图；最后结合激光传感器的深度数据，采用抽样一致性算法，计算深度标尺，并使用对透视平面进行深度赋值，将相对深度图转化为绝对深度图，进而完成对透视平面的深度预测。实验结果表明该算法能成功检测并分割透视平面，且能得到正确的透视平面绝对深度信息。

Abstract:

Aiming at the problem that a single laser sensor or vision sensor cannot detect the perspective three-dimensional plane, a perspective plane detection and depth prediction algorithm based on the fusion of laser sensor and vision sensor is proposed. Firstly, GDNet is used to segment the perspective plane in a two-dimensional color image; secondly, a single image reflection removal algorithm is applied to separate the background information in the segmented perspective plane area, and the MegaDepth algorithm is used for depth prediction to obtain a relative depth map; and finally Combining the depth data of the laser sensor, the sampling consistency algorithm is used to calculate the depth scale, and the depth assignment to the perspective plane is used to convert the relative depth map into an absolute depth map, and then complete the depth prediction of the perspective plane. The experimental results show that the algorithm can successfully detect and segment the perspective plane, and can obtain the correct absolute depth information of the perspective plane.