This article introduces a novel feature-based visual simultaneous localization and mapping (v-SLAM), termed finite-plane (FP)-SLAM, using an RGB-D camera Specifically, the problem solved in this article is to extract sparse features effectively from RGB-D images and build a graph consisting of poses and sparse features as accurately as possible in real time on CPU. Motivated by surfel-point SLAM (SP-SLAM), FP-SLAM extracts points and surfels from RGB-D images as sparse features and optimizes a pose and feature graph jointly. Compared to SP-SLAM, however, two new residuals are proposed to optimize the graph and improved the accuracy in FP-SLAM. These new residuals exploit the relationships between surfel-surfel (SS) and between point-surfel (PS), capturing interfeature relations. Incorporating these new residuals with those from SP-SLAM, a total of four residuals are employed to jointly optimize points, surfels, and camera poses. In addition, to expedite processing, the Jacobian of the error function is directly implemented in the optimization instead of using the off-the-shelve derivative module to speed up the processing. Finally, the effectiveness of the proposed FP-SLAM is validated on benchmark datasets by comparing it against previous methods in terms of localization accuracy.

In this paper, we propose a novel image retrieval network named Correlation Verification Network (CVNet) to replace the conventional geometric re-ranking with a 4D convolutional neural network that learns diverse geometric matching possibilities. To enable efficient cross-scale matching, we construct feature pyramids and establish cross-scale feature correlations in a single inference, thereby replacing the costly multi-scale inference. Additionally, we employ curriculum learning with the Hide-and-Seek strategy to handle challenging samples. Our proposed CVNet demonstrates state-of-the-art performance on several image retrieval benchmarks by a large margin. From an implementation perspective, however, CVNet has one drawback: it requires high memory usage because it needs to store dense features of all database images. This high memory requirement can be a significant limitation in practical applications. To address this issue, we introduce an extension of CVNet called Dense-to-Sparse CVNet (CVNet), which can significantly reduce memory usage by sparsifying the features of the database images. The sparsification module in CVNet learns to select the relevant parts of image features end-to-end using a Gumbel estimator. Since the sparsification is performed offline, CVNet does not increase online extraction and matching times. CVNet dramatically reduces the memory footprint while preserving performance levels nearly identical to CVNet.

Three-dimensional light detection and ranging (3D LiDAR) sensors are widely used in autonomous vehicles. Among various perception problems using a 3D LiDAR, the identification of drivable region (DR) is one of the most important problems. In this paper, the DR identification using a 3D LiDAR is reformulated as a completion problem. In other words, the problem is to generate the virtual points so that the virtual points cover the DR as densely and uniformly as possible, thereby representing the DR well enough. The problem considered herein is named the DR completion problem. To solve the DRC, a new network called the Drivable Region Completion Network (DRCN) is proposed. The proposed DRCN consists of an encoder part and a decoder part. The encoder part consists of two encoders. The first one is the Multi-Layer Perceptron (MLP) encoder, and it captures the global feature of the given LiDAR points. The second one is named the Point Pyramid Network (PPN) encoder, and it extracts local features between the points in the point cloud. The decoder part takes the coarse-to-fine structure, and it consists of the coarse and refinement decoders. The coarse decoder predicts seed points for the DR. The refinement decoder refines the seed points to predict the final dense DR points. Finally, the DRCN is applied to the SemanticKITTI dataset for evaluation. The proposed DRCN is validated by showing that our DRCN outperforms other DR detection methods in terms of accuracy.

This article introduces a novel feature-based visual simultaneous localization and mapping (v-SLAM), termed finite-plane (FP)-SLAM, using an RGB-D camera Specifically, the problem solved in this article is to extract sparse features effectively from RGB-D images and build a graph consisting of poses and sparse features as accurately as possible in real time on CPU. Motivated by surfel-point SLAM (SP-SLAM), FP-SLAM extracts points and surfels from RGB-D images as sparse features and optimizes a pose and feature graph jointly. Compared to SP-SLAM, however, two new residuals are proposed to optimize the graph and improved the accuracy in FP-SLAM. These new residuals exploit the relationships between surfel-surfel (SS) and between point-surfel (PS), capturing interfeature relations. Incorporating these new residuals with those from SP-SLAM, a total of four residuals are employed to jointly optimize points, surfels, and camera poses. In addition, to expedite processing, the Jacobian of the error function is directly implemented in the optimization instead of using the off-the-shelve derivative module to speed up the processing. Finally, the effectiveness of the proposed FP-SLAM is validated on benchmark datasets by comparing it against previous methods in terms of localization accuracy.

In this paper, we propose a novel image retrieval network named Correlation Verification Network (CVNet) to replace the conventional geometric re-ranking with a 4D convolutional neural network that learns diverse geometric matching possibilities. To enable efficient cross-scale matching, we construct feature pyramids and establish cross-scale feature correlations in a single inference, thereby replacing the costly multi-scale inference. Additionally, we employ curriculum learning with the Hide-and-Seek strategy to handle challenging samples. Our proposed CVNet demonstrates state-of-the-art performance on several image retrieval benchmarks by a large margin. From an implementation perspective, however, CVNet has one drawback: it requires high memory usage because it needs to store dense features of all database images. This high memory requirement can be a significant limitation in practical applications. To address this issue, we introduce an extension of CVNet called Dense-to-Sparse CVNet (CVNet), which can significantly reduce memory usage by sparsifying the features of the database images. The sparsification module in CVNet learns to select the relevant parts of image features end-to-end using a Gumbel estimator. Since the sparsification is performed offline, CVNet does not increase online extraction and matching times. CVNet dramatically reduces the memory footprint while preserving performance levels nearly identical to CVNet.

Three-dimensional light detection and ranging (3D LiDAR) sensors are widely used in autonomous vehicles. Among various perception problems using a 3D LiDAR, the identification of drivable region (DR) is one of the most important problems. In this paper, the DR identification using a 3D LiDAR is reformulated as a completion problem. In other words, the problem is to generate the virtual points so that the virtual points cover the DR as densely and uniformly as possible, thereby representing the DR well enough. The problem considered herein is named the DR completion problem. To solve the DRC, a new network called the Drivable Region Completion Network (DRCN) is proposed. The proposed DRCN consists of an encoder part and a decoder part. The encoder part consists of two encoders. The first one is the Multi-Layer Perceptron (MLP) encoder, and it captures the global feature of the given LiDAR points. The second one is named the Point Pyramid Network (PPN) encoder, and it extracts local features between the points in the point cloud. The decoder part takes the coarse-to-fine structure, and it consists of the coarse and refinement decoders. The coarse decoder predicts seed points for the DR. The refinement decoder refines the seed points to predict the final dense DR points. Finally, the DRCN is applied to the SemanticKITTI dataset for evaluation. The proposed DRCN is validated by showing that our DRCN outperforms other DR detection methods in terms of accuracy.

on DAVIS 2016 validation set are 0.12 and 0.13 seconds per frame, respectively, and the two networks have similar computation times to STM.

In this paper, we present a novel geometry-aware RGB-D Gaussian Splatting SLAM system, named G2S-ICP SLAM. The proposed method performs high-fidelity 3D reconstruction and robust camera pose tracking in real-time by representing each scene element using a Gaussian distribution constrained to the local tangent plane. This effectively models the local surface as a 2D Gaussian disk aligned with the underlying geometry, leading to more consistent depth interpretation across multiple viewpoints compared to conventional 3D ellipsoid-based representations with isotropic uncertainty. To integrate this representation into the SLAM pipeline, we embed the surface-aligned Gaussian disks into a Generalized ICP framework by introducing anisotropic covariance prior without altering the underlying registration formulation. Furthermore we propose a geometry-aware loss that supervises photometric, depth, and normal consistency. Our system achieves real-time operation while preserving both visual and geometric fidelity. Extensive experiments on the Replica and TUM-RGBD datasets demonstrate that G2S-ICP SLAM outperforms prior SLAM systems in terms of localization accuracy, reconstruction completeness, while maintaining the rendering quality.

We present a novel, training-free approach to scene change detection. Our method leverages tracking models, which inherently perform change detection between consecutive frames of video by identifying common objects and detecting new or missing objects. Specifically, our method takes advantage of the change detection effect of the tracking model by inputting reference and query images instead of consecutive frames. Furthermore, we focus on the content gap and style gap between two input images in change detection, and address both issues by proposing adaptive content threshold and style bridging layers, respectively. Finally, we extend our approach to video, leveraging rich temporal information to enhance the performance of scene change detection. We compare our approach and baseline through various experiments. While existing train-based baseline tend to specialize only in the trained domain, our method shows consistent performance across various domains, proving the competitiveness of our approach.

score of 73.9%, improving by 32.9% over the baseline. The HTC600 dataset used in our experiments is publicly available to support autonomous driving research.