Research And Application Of SLAM Backend Optimization Based On G2o

With the rapidly development of computer vision,many related technologies have been applied to the field of robot.SLAM is considered to be the basis for the full autonomy of mobile robots.Graph-based SLAM is mainly into front end and back end.The front end mainly obtains the initial pose of the robot and the coordinates of space landmark based on sensor measurement.The back end is mainly to optimize the date from the front end which contains measurement noise.The optimization methods of SLAM back-end are mainly divided into two categories,one is based on filtering optimization method,and the other is graph-based optimization method,in which graph-based optimization method has become the mainstream of current research.we can know that the filtering-based optimization method only estimates the current state,and it will generate cumulative error.Not only does it cannot guarantee the consistency and accuracy of the optimized results,but it is not suitable for large-scale environments.The graph-based optimization method performs one-time processing on the information of all the moments of the mobile robot,so the consistency and accuracy of the obtained optimization results are higher than the filtering method.This paper first analyzes two different graph optimization methods based on BA and based on posegraph,and then uses the g2 o framework to design two different graph optimization system.For the shortcomings of the LM algorithm used in the optimization process,put forward an LM algorithm of changing the trust region.The algorithm newly defines the non-negative parameters ?.It Controls the step size of by changing the size of the trust region in the iterative process.Meanwhile,it can guarantee the square convergence of the algorithm.Finally,we through experiments to analyze BA SLAM and posegraph SLAM and compare experimental results between the improved LM algorithm and LM algorithm.According to the absolute pose error ATE and relative pose error RPE and the relation chart between the total error and the number of iterations.We can that the improved LM algorithm not only improves the accuracy of the robot's motion trajectory,but also improves the iterative efficiency in the optimization process.