Inexpensive Stereo System Using ReStereoNet for Disparity Map Estimation

摘要

The article presents an idea of developing an inexpensive stereo camera system to replace for 3D camera on mobile robots. The stereo system is built from two in-expensive camera installed in an 3D printed base. To gain disparity map from left and right images, the system applies PSMNET algorithm, a Convolutional Neural Network based method. The system with PSMNET performs better than traditional block matching algorithm in term of accuracy and robustness. The experimental results show possibility of using that camera system on mobile robot for autonomous navigation tasks.