Improved noise-adapted semantic SLAM

Based on the rapid development of deep learning, semantic information has gradually become a research hotspot in the field of SLAM (Simultaneous Location and Mapping). The noise problem caused by the environment and sensor results in the lack of consistency of semantic maps, and affects the accuracy of the algorithms. Loss function can adjust the weights assigned to the outliers, so it can reduce the impact of the outliers. However, the model of loss function used by most semantic SLAM is fixed and cannot adapt well to the changing environment. To solve this problem, this paper proposes a improved noise-adapted semantic SLAM, which uses Gaussian mixture correntropy weight function as loss function. Its model structure is variable by adjusting the parameters in changing environment, so it can adapte the noise distribution to the greatest extent, which is more conducive to reducing the weight of the algorithm for outliers and improving robustness to the outliers. Experiments on the public KITTI dataset show that the average relative translation and rotation error of the proposed method are reduced by 4.08% and 5.55%, the constructed semantic maps are more consistent.