Multi-camera System Depth Estimation

Abstract

For the problem of multi-camera image depth measurement, a multi-camera depth estimation system based on semantic segmentation is proposed in this paper. The system first obtains the depth map between neighboring cameras by multi-camera calibration and stereo matching, then stitches multi-camera images by feature extraction and matching, then uses semantic segmentation network to identify and segment objects, and establishes an object height library, and finally estimates the object distance in the non-overlapping region based on the similar triangle principle and the object height library. The object depth obtained by the method of this paper has certain reference value and basically meets the distance measurement requirements of SLAM.