Monocular Visual Odometry Based on Plane Tracking with Rectangular Constraint

This paper presents a feature-based monocular visual odometry method by detecting and tracking plane of man-made buildings. We utilize the vertical and horizontal structural lines on a plane and rectangular constraint to reconstruct a mesh plane in 3-D space from single 2-D image, tracking the target plane and estimating 6-degree of freedom poses through monocular RGB camera. A reliability testing is integrated, which can reduce the impact of drift and speed up the process through downsampling the input frames. We have evaluated our method through simulation and experiment, The results indicate that our method performs well in noisy environments and can operate in real-time.