RC-SLAM: Road Constrained Stereo Visual SLAM System Based on Graph Optimization

Intelligent vehicles are constrained by road, resulting in a disparity between the assumed six degrees of freedom (DoF) motion within the Visual Simultaneous Localization and Mapping (SLAM) system and the approximate planar motion of vehicles in local areas, inevitably causing additional pose estimation errors. To address this problem, a stereo Visual SLAM system with road constraints based on graph optimization is proposed, called RC-SLAM. Addressing the challenge of representing roads parametrically, a novel method is proposed to approximate local roads as discrete planes and extract parameters of local road planes (LRPs) using homography. Unlike conventional methods, constraints between the vehicle and LRPs are established, effectively mitigating errors arising from assumed six DoF motion in the system. Furthermore, to avoid the impact of depth uncertainty in road features, epipolar constraints are employed to estimate rotation by minimizing the distance between road feature points and epipolar lines, robust rotation estimation is achieved despite depth uncertainties. Notably, a distinctive nonlinear optimization model based on graph optimization is presented, jointly optimizing the poses of vehicle trajectories, LPRs, and map points. The experiments on two datasets demonstrate that the proposed system achieved more accurate estimations of vehicle trajectories by introducing constraints between the vehicle and LRPs. The experiments on a real-world dataset further validate the effectiveness of the proposed system.

A Novel Classification Framework for Hyperspectral Image Data by Improved Multilayer Perceptron Combined with Residual Network

Convolutional neural networks (CNNs) have attracted extensive attention in the field of modern remote sensing image processing and show outstanding performance in hyperspectral image (HSI) classification. Nevertheless, some hyperspectral images have fixed position priors and parameter sharing between different positions, so the common convolution layer may ignore some important fine and useful information and cannot guarantee to effectively capture the optimal image features. This paper proposes an improved multilayer perceptron (IMLP) and IMLP combined with ResNet (IMLP-ResNet) two models for HSI classification. Combined with the characteristics of hyperspectral data, we design IMLP based on three improvements. Specifically, a depthwise over-parameterized convolutional layer is introduced to increase learnable parameters of the model in IMLP, which speeds up the convergence of the model without increasing the computational complexity. Secondly, a Focal Loss function is used to suppress the useless ones in the classification task and enhance the critical spectral–spatial features, which allow the IMLP network to learn more useful hyperspectral image information. Furthermore, to enhance the convergence speed of the network, cosine annealing is introduced to further improve the training performance of IMLP. Furthermore, the IMLP module is combined with a residual network (IMLP-ResNet) to construct a symmetric structure, which extracts more advanced semantic information from hyperspectral images. The proposed IMLP and IMLP-ResNet are tested on the two public HSI datasets (i.e., Indian Pines and Pavia University) and a real hyperspectral dataset (Xuzhou). Experimental results demonstrate the superiority of the proposed IMLP-ResNet method over several state-of-the-art methods with the highest OA, which outperforms CNN by 8.19%, 6.28%, 5.59% and outperforms ResNet by 3.52%, 3.54%, 2.67% on Indian Pines, Pavia University and Xuzhou datasets, respectively, and demonstrates that the well-designed MLPs can also obtain remarkable classification performance of HSI.