SLAM localizing methods based on camera?

edit

You are not able to apply Gmapping directly for this application because it does not give a promising result when enough features are not available in the sensor field of view. RGB-D SLAM also will not work as you have repetitive patterns on your RGB images.

I think your problem seems easy than general SLAM problem. You can extract those line segments using a line detector from RGB image. Then use ICP algorithm to align those frames.

If you need a more robust result, you can use the same method with Rao-Blackwellized particle filter that is used in Gmapping. But you need an accurate data association strategy to work with RBPF.