Authors : Mohamed H. Mahmoud, Nashaat Hissen and Elsayed Hemayed
Abstract: Recently graph based SLAM has become a popular representation for solving SLAM problem but it is affected by its inherent nonlinearity due to noisy measurements and reprojection errors which lead to non-consistent map. The proposed system provides the entire robot trajectory at lower cost using stereo camera, generates a globally consistent 3D map and provides linear graph optimization. We show that the validation of our system based on stereo camera which utilizes a linear optimization method which provides only one accurate solution in contrast to the well-known nonlinear global optimization like BA of ORB-SLAM which provides multiple non-accurate solutions for optimization. Furthermore, our system runs in real time and is 3X faster than the current state-of-the-art SLAM systems for stereo SLAM. Our system gives comparable accuracy compared with the state of the art. We tested our system using KITTI data sets. We show that our system can be an alternative to bundle adjustment approaches with a better accuracy, robustness and more efficient than some well-known systems.
Mohamed H. Mahmoud, Nashaat Hissen and Elsayed Hemayed, 2019. A Linearly Optimization Method for Slam Estimation Based on Stereo Features. Agricultural Journal, 14: 109-117.