Urban streets form the backbone of cities, and variations in street-junction spatial configuration provide valuable insights into urban form and function, both of which exhibit fractal and scaling characteristics in their structural and statistical properties. However, these patterns emerge from the interaction of multiple spatial processes operating at different scales, making it difficult to capture and compare their form-function relationships using single-scale analyses. To address this challenge, this study proposes a new α–β framework for multi-scale urban form comparison, grounded in fractal theory. Using a progressive clustering method, we construct scaling profiles for multiple cities, deriving the α (form) and β (function) parameters to systematically compare urban structures with differing morphologies and to identify urban clusters at their characteristic scales. This approach transforms the subjective perception of morphology into an actionable analytical tool, enabling integrated form-function assessment and overcoming the challenges of comparative studies from different urban systems. Overall, the proposed α–β framework effectively deconstructs urban complexity, as evidenced by its application to a comparative analysis of nine global metropolises.

Robust visual place recognition (VPR) is fundamental to autonomous navigation systems. However, conventional two-stage pipelines often suffer from a severe accuracy-efficiency trade-off, achieving high performance only at the expense of computationally intensive local matching. To address this limitation, we propose a unified framework that integrates global retrieval and re-ranking within a single Vision Transformer (ViT) backbone. During training, we introduce an age-weighted memory queue with exponential decay to facilitate large-scale hard-negative mining. This mechanism effectively mitigates feature drift and enhances the discriminative power of global descriptors. During inference, we leverage the ViT's inherent self-attention and value representations to perform salient region localization and local matching. This attention-guided re-ranking strategy requires no additional parameters or training, yet effectively corrects initial retrieval errors. Extensive experiments demonstrate that our method improves Recall@1 by up to 7.2% under challenging environmental conditions, such as nighttime and rainy weather. Furthermore, it achieves a 12 & times; inference speedup and reduces storage overhead by 26.3% compared to state-of-the-art two-stage approaches. Our contributions include: (1) empirical evidence that age-based weighting in the memory queue minimizes gradient bias induced by feature drift, thereby stabilizing training; and (2) a demonstration that a ViT's internal representations can be decoupled to enable near-zero overhead re-ranking, offering a highly efficient alternative for robust VPR.

Urban mobility is shaped by the coupled effects of regional interaction demand, network-constrained route realization, and activity-purpose context. However, existing flow-clustering studies often examine these dimensions separately: OD-based methods reveal macro spatial connectivity but suppress route-level heterogeneity, whereas path-level trajectory methods capture movement similarity but are rarely linked back to multiscale interaction structures and semantic travel contexts. This study proposes a macro-micro-coupled flow-clustering framework to bridge this methodological gap and provide a multiscale representation of heterogeneous urban mobility dynamics. The framework constructs OD flows and map-matched path flows from trajectory data, identifies macro interaction structures using scale-aware OD clustering, and develops a movement-consistent path-flow clustering procedure to characterize network-constrained route realizations. Inferred trip-purpose labels are further used to semantically stratify topological path-flow clusters, enabling mobility heterogeneity to be interpreted across spatial, topological, and activity-context dimensions. Using 1.39 million ride-hailing trips in Wuhan, China, the path-flow clustering stage achieved high retained-cluster route-realization quality, with representative-path fidelity (RPF) of 0.9669, turning-signature consistency (TSC) of 0.8784, and Fowlkes-Mallows index (FMI) of 0.6393, exceeding the controlled comparator settings in shared-input diagnostics. The coupled framework further revealed path-coherence differences across trip distances, route-overlap bottlenecks, congestion-exposure heterogeneity, and trip-purpose-specific route response patterns under observed congestion exposure, supporting activity-context-aware interpretation of multiscale urban mobility heterogeneity.

Visual Place Recognition (VPR) is pivotal for robust localization in autonomous navigation, yet long-term environmental dynamics and resource-intensive video models challenge its efficacy. This work introduces a parameter-efficient image-to-video transfer learning framework (I2VPR) to address these limitations, leveraging abundant image data to enhance video-based VPR. We propose a novel spatio-temporal convolution adapter (ST-ConvAdapter), a lightweight plug-and-play module that, unlike prior adapters which rely on MLPs or 2D convolutions, is specifically designed to model inter-frame dynamics. Integrated into pre-trained vision transformers, it enables hierarchical spatio-temporal feature learning with minimal parameter overhead. Leveraging depthwise 3D convolutions, I2VPR transfers robust spatial features from image models to video, effectively bridging domain gaps, suppressing noise, and capturing motion dynamics. Evaluations on benchmark datasets demonstrate that I2VPR surpasses state-of-the-art methods, achieving a 94.1% Recall@1 in cross-domain tests — a 15% improvement over prior art. Ablation studies confirm the design efficacy, highlighting its ability to balance spatial invariance and temporal sensitivity. This framework offers an efficient, scalable solution for resource-constrained VPR systems in dynamic real-world settings.

Modern indoor spaces increasingly contain curved walls, slanted surfaces, nested rooms and other non-Manhattan structures, making room-level subdivision from 3D point clouds challenging. Existing projection-based, primitive-based and semantic methods often rely on Manhattan assumptions, explicit structural labels or local geometric heuristics, which may lead to fragmented spatial units and unstable boundaries in complex scenes. Here, we propose a topology-driven voxel partitioning framework that reformulates indoor space subdivision as controlled connectivity disconnection within a topological closure. The central idea is to first construct a closed voxelated space domain and then selectively disconnect it at near functional openings and topological bottlenecks, rather than partitioning space only from local geometric cues. Within this framework, classical operations are reorganized under topological constraints as follows: adaptive region growing with aperture-sensitive spherical kernels generates initial spatial units, boundary-anisotropic watershed completion restores unlabeled boundary regions within interior domain, and label reassignment regularizes shared interfaces by minimizing discrete contact areas. Experiments on real-world and synthetic datasets show that the method produces stable room-level subdivisions across most Manhattan and non-Manhattan scenes. Cases with narrow bottlenecks or abrupt geometric narrowing still reduce detection-level precision and recall, indicating remaining limitations in highly constrained spatial configurations. Overall, the proposed framework offers a useful topological modeling perspective for indoor space subdivision in complex non-Manhattan environments.

This study conducts a comparative analysis of machine learning algorithms for predicting housing prices in major Chinese cities, addressing the growing complexity and economic significance of the real estate market. With housing affordability becoming a critical socioeconomic issue, accurate price prediction models are essential for policymakers, investors, and urban planners. The research evaluates the performance of four algorithms—Linear Regression, Decision Trees, Random Forests, and Gradient Boosting Machines—using a dataset comprising historical transaction records, macroeconomic indicators, and location-based features from 10 major Chinese cities. Results indicate that ensemble methods, particularly Gradient Boosting Machines, achieve the highest predictive accuracy, with a mean absolute error reduction of up to 18% compared to traditional linear models. The findings underscore the importance of incorporating non-linear relationships and feature interactions in housing price modeling. This study not only provides a practical framework for real estate valuation but also highlights the potential of advanced machine learning techniques in addressing urban economic challenges.

A digital elevation model (DEM) is widely recognized as the most effective digital representation of the Earth’s surface and serves as the fundamental platform for simulating various Earth systems. Extensive efforts have been devoted to exploring methods for generating high-fidelity DEM datasets that are computationally efficient for diverse applications. However, the existing methods do not guarantee the optimal digital representation of the Earth’s surface. This study proposed a novel curvature-based geodesic centroidal Voronoi tessellation method for generating a topographic triangulated irregular network (TIN) DEM based on optimal transport theory. This study is the first to present a globally optimized digital representation of the Earth’s surface with a predetermined number of vertices, which is crucial for computational feasibility. This study achieves the optimal TIN by measuring mean curvature and introducing geodesic distances on the topographic surface. Representative vertices that best adapt to the topography are identified through an optimal surface approximation process. Experimental results confirm that the proposed method effectively generates the optimal digital representation of the topographic surface with the lowest elevation errors and minimal deviations from the original topographic features. By generating optimal TIN DEM with any desired number of vertices, the proposed method not only balances high-precision representation and computational efficiency but also offers a novel approach to deepening the understanding of topographic structures. Furthermore, it provides an effective solution for compressing extensive topographic data and facilitating multiscale representation of the Earth’s surface.

Driver gaze estimation task is pivotal for safe driving. However, challenges persist when dealing with changing illumination, eyeglasses, adjacent zones, or personal behavior and appearance. To tackle these problems, we introduce a driver gaze estimation approach, including gaze zone and direction estimation. We propose a global facial feature extraction convolutional neural network (gCNN) embedded with attention network for driver gaze zone estimation. The incorporation of attention mechanisms in different dimensions (channel or spatial) at various stages facilitates the network in efficiently capturing overall generic features in early stages and concrete representations in later stages. This network is also applied to extract facial features in gaze direction estimation task. While a local eye feature extraction convolutional neural network (LeCNN) is proposed for fine-grained eye features extraction. The facial and eye features, as well as head pose, are concatenated and fused to regress the finer gaze direction. The experimental results show that the network achieves an error of 2.43 degrees and 4.36 degrees on MPIIFaceGaze and EyeDIAP datasets, respectively, outperforming the prior arts. Furthermore, in driver gaze zone estimation task, our method achieves accuracy of 98.87 % on Laboratory for Intelligent and Safe Automobiles (LISA) Gaze dataset, with 3.91 % improvement over prior arts. It also achieves a compatible performance of 82.80 % on Driver Gaze in the Wild (DGW) dataset.

A significant challenge remains in identifying and modeling complex 3D point cloud indoor scenes, especially when indoor structural points exhibit similar spatial geometry. By combining local point convolution and global sparse vector attention (SVA), we propose a network to improve the accuracy of semantic segmentation for challenging and often ambiguous wall surfaces and neighboring components with similar structural geometry. Specifically, maximum priori attention (MPA) enhances focus on critical points to optimize feature aggregation in local multiscale multilayer perception (MLP). Then we propose a PCA-based 3D projective point convolution (PPC), capable of orientating the principal direction of the neighborhood and identifying the projection plane. And the neighborhood points are projected adaptively as convolution kernel points, which can be adjusted dynamically to accommodate neighborhood geometric variations and differences. A more flexible convolution whose weights are learned using the projection kernel points is then executed to emphasize the primary structural features and accommodate geometric variations. Globally, SVA effectively models the interdependence and spatial constraint between distant points. The weighted addition vector representation can adaptively adjust the feature channel weight across different dimensions, highlighting critical features while reducing the loss of fine details caused by scalarized compression. By focusing only on relevant points, sparse attention significantly reduces computation burden without sacrificing accuracy. Experimental results show that our proposed model achieves over 73% mean intersection over union (mIoU) on the S3DIS dataset, maintaining excellent recognition performance for ceilings and floors while significantly improving the segmentation of walls and adjacent structures such as doors, windows, boards, and bookcases, which are commonly prone to confusion. These results confirm the model’s robustness and effectiveness in handling complex indoor environments.

With the rapid growth of online education platforms, accurately predicting volatile and nonlinear course resource traffic remains challenging due to the limitations of traditional time-series models like ARIMA in capturing spatiotemporal dependencies. This study proposes a hybrid deep learning model integrating Long Short-Term Memory (LSTM) networks and Graph Neural Networks (GNN) to address this issue. The LSTM component models temporal patterns in historical access data, while the GNN captures spatial interactions between course resources through a dynamically constructed knowledge graph based on multidimensional associations (e.g., categories, temporal correlations, and semantic similarities). A weighted fusion strategy combines temporal and spatial features to enhance prediction robustness. Experimental results demonstrate that the hybrid model achieves superior performance over standalone LSTM and ARIMA, with 0.40 MSE and 0.35 MAE, and exhibits strong generalization across holiday scenarios and device-specific data. The approach effectively balances accuracy and stability, offering a scalable solution for optimizing resource allocation in online education platforms.

In road scenes, the collected 3D point cloud data is usually accompanied by a large amount of interference mainly composed of ground point clouds and the property of uneven density distribution, which will bring difficulties to subsequent recognition and prediction. To address these problems, this paper proposes a ground point cloud segmentation algorithm based on local density plane fitting. Firstly, for the uneven density distribution of 3D point clouds, density segmentation is used to obtain several regions with balanced density. Then, candidate sample selection and plane validity detection are carried out for each region. The modified classical DBSCAN clustering algorithm is used to obtain effective fitting planes and perform clustering according to the fitting planes. Finally, different planes are divided according to the clustering results, and abnormal inspection is performed on the obtained results to screen out the most reasonable result. This scheme can effectively improve the scalability of the algorithm, reduce training costs, and improve deployment efficiency and universality. Experimental results show that the algorithm used in this paper has advantages compared with advanced algorithms of the same category, and can greatly reduce ground interference.

Health inequality induced by environmental degradation has become a growing global concern. Especially for migrants across different medical insurance pooling regions, the problem is even more serious. Drawing on micro-level data from the China Health and Retirement Longitudinal Survey (CHARLS) and official air pollution data, we employ a two-stage least squares (2SLS) estimation strategy that adopts a relative deprivation perspective to identify causal effects. The findings suggest that air pollution significantly worsens health inequality, particularly among women, the elderly, and socioeconomically disadvantaged groups. Furthermore, the analysis reveals that Cross-regional Medical Insurance Scheme(CRMIS) plays a mitigating role in reducing inequality caused by air pollution. For migrants, choosing the medical insurance in the inflow region can alleviate the severity of health inequalities caused by air pollution. This research contributes to the global literature on environmental inequality by highlighting how institutional mechanisms, such as portable health insurance, can buffer the unequal health impacts of pollution. The insights derived from China’s experience offer valuable policy implications for other countries facing rapid urbanization, internal migration, and rising environmental health risks.

Download This Paper Open PDF in Browser Add Paper to My Library Share: Permalink Using these links will ensure access to this page indefinitely Copy URL Unscented Lamarckian Kalman Filter for Nonlinear Estimation and Target Tracking 12 Pages Posted: 23 Feb 2024 See all articles by Lin LiLin LiQingdao Agricultural UniversityJiang Xiaoaffiliation not provided to SSRNYun BaiChongqing Technology and Business UniversityYun Liaffiliation not provided to SSRN Abstract In object tracking, autonomous driving, robotics, and artificial intelligence (AI), the Particle Filter (PF) has emerged as a preferred filtering solution for nonlinear and non-Gaussian systems. Nonetheless, the PF faces challenges such as complexity, particle degeneracy and impoverishment. This study first provides an exhaustive exploration of the PF mechanism, alongside associated evolutionary principles. Then, it develops a Lamarckian Inheritance Resampling (LIR) mechanism, intricately designed for the PF architecture and leveraging on AI principles. This culminates in the novel Unscented Lamarckian PF (ULPF), integrating LIR with Unscented Kalman Filtering (UKF). For the ULPF, Lamarckian evolution enhances particle diversity and UKF combats particle degeneracy. Empirical evaluations verify ULPF's superior performance over traditional PF in nonlinear estimation. Compared to the state-of-the-art PF via Differential Evolution, ULPF achieves reduced computational complexity, faster filtering, and greater precision. These attributes underscore ULPF's efficacy in approximating posterior distribution, making it especially suitable for targets tracking. We further demonstrate the algorithm performance through an example of two active sensors tracking a target for autonomous driving and thus for potentially more applications to dynamic vision sensing. Keywords: Particle filter, Lamarckian evolutionary principles, Lamarckian inheritance resampling, unscented Lamarckian PF, unscented Kalman Filter, Target-tracking Suggested Citation: Suggested Citation Li, Lin and Xiao, Jiang and Bai, Yun and Li, Yun, Unscented Lamarckian Kalman Filter for Nonlinear Estimation and Target Tracking. Available at SSRN: https://ssrn.com/abstract=4736991 Lin Li Qingdao Agricultural University ( email ) China Jiang Xiao affiliation not provided to SSRN ( email ) No Address Available Yun Bai Chongqing Technology and Business University ( email ) Chongqing, 400067China Yun Li (Contact Author) affiliation not provided to SSRN ( email ) No Address Available Download This Paper Open PDF in Browser Do you have negative results from your research you’d like to share? Submit Negative Results Paper statistics Downloads 0 Abstract Views 11 38 References PlumX Metrics Related eJournals Applied Computing eJournal Follow Applied Computing eJournal Subscribe to this fee journal for more curated articles on this topic FOLLOWERS 229 PAPERS 7,432 Feedback Feedback to SSRN Feedback (required) Email (required) Submit If you need immediate assistance, call 877-SSRNHelp (877 777 6435) in the United States, or +1 212 448 2500 outside of the United States, 8:30AM to 6:00PM U.S. Eastern, Monday - Friday.

Complex nested and discontinuous location references are common in unstructured text. Extracting them is essential for accurate location information retrieval and spatial inference. However, traditional methods struggle with these references due to annotation system and model architecture limitations. In this study, we introduce a deep learning approach to uniformly recognize flat, nested, and discontinuous location references, motivated by recognizing fine-grained expressway location references. The approach uses a pre-trained language model to generate semantic sentence representations and a distance and direction-aware Transformer for contextual encoding. Then, it recognizes location references by modeling the adjacency and boundary relations between word pairs. We evaluated the approach on seven benchmark datasets and compared it with state-of-the-art methods. The results show that the approach achieves higher accuracy with faster inference, validating our modeling paradigm and architecture. The ablation study further confirms the effectiveness of submodules in architecture. These findings can provide valuable insights for developing advanced unified location reference recognition methods. Moreover, the detailed labeled dataset for location references can facilitate the evaluation and comparison of unified recognition methods and systems.

In this paper, we propose an effective method for mismatch removal, termed as graph neighborhood motion consensus, to address the feature matching problem which plays a pivotal role in various computer vision tasks. In our method, we convert each feature correspondence into a motion field sample and model it with the probabilistic graphical model (PGM). To differentiate mismatches from true matches, we firstly design a metric based on neighborhood topology consensus and neighborhood interaction to evaluate the correctness of each match. We also design a variance-based similarity search module to make the information used more reliable for better matching performance. To derive the solution of PGM, we build a model to transform the problem into an integer quadratic programming problem and obtain its closed-form solution with linear time complexity. Extensive experiments on general feature matching, fundamental matrix estimation and image registration tasks demonstrate that our proposed method can achieve superior performance over several state-of-the-art approaches.

Ecological security patterns (ESPs) provide an effective spatial approach for identifying critical conservation areas and ensuring regional ecological security. However, prior research has not paid much attention to the importance of the stability of ecological sources in time-series changes, which is especially critical for maintaining ecological functions in ecologically fragile areas. Focusing on the Ningxia Hui Autonomous Region (NHAR) of China, this study evaluated the spatiotemporal change patterns in ecosystem services importance (ESI) from 2000 to 2020, integrating the spatial principal component analysis (SPCA) and circuit theory to propose a novel ESP construction framework that aims to address the issue of insufficient consideration of source stability. A total of 93 stable ecological sources were identified, with the capacity to ensure the continuous provision of high-level ecosystem services and resistance to external disturbances. The extraction of 234 ecological corridors and 430 ecological nodes effectively enhanced the stable flow of ecological processes and connectivity. The stable ESP, constituted by the above ecological elements, can serve as core ecological space and basic skeleton to maintain the regional sustainable landscape. This study provides scientific references for identifying key priority conservation areas and formulating targeted ecological conservation and restoration strategies in ecologically fragile areas.

Gaze estimation, which has a wide range of applications in many scenarios, is a challenging task due to various unconstrained conditions. As information from both full-face and eye images is instrumental in improving gaze estimation, many multiregion gaze estimation models have been proposed in recent studies. However, most of them simply use the same regression method on both eye and face images, overlooking that the eye region may contribute more fine-grained features than the full-face region, and the variation in the left and right eyes of an individual caused by head pose, illumination, and partially occluded eye may lead to inconsistent estimations. To address these issues, we propose an appearance-based end-to-end learning network architecture with an attention mechanism, named efficient gaze network (EG-Net), which employs a two-branch network for gaze estimation. Specifically, a base CNN is utilized for full-face images, while an efficient eye network (EE-Net), which is scaled up from the base CNN, is used for left- and right-eye images. EE-Net uniformly scales up the depth, width and resolution of the base CNN with a set of constant coefficients for eye feature extraction and adaptively weights the left- and right-eye images via an attention network according to its "image quality". Finally, features from the full-face image, two individual eye images and head pose vectors are fused to regress the eye gaze vectors. We evaluate our approach on 3 public datasets, the proposed EG-Net model achieves much better performance. In particular, our EG-Net-v4 model outperforms state-of-the-art approaches on the MPIIFaceGaze dataset, with prediction errors of 2.41 cm and 2.76 degrees in 2D and 3D gaze estimation, respectively. It also yields a performance improvement to 1.58 cm on GazeCapture and 4.55 degrees on EyeDIAP dataset, with 23.4 % and 14.2 % improvement over prior arts on the two datasets respectively. The code related to this project is open-source and available at https://github.com/wuxinmei/EE_Net.git.

The spatial balance of health insurance is an important prerequisite for social security equity, so analyzing the spatial differences in the development level of Urban Employee Basic Health Insurance (BHI) can help promote regional health equity. Based on the data of BHI in 31 provinces and cities in China from 2003 to 2020, this paper explores the spatial difference, dynamic evolution and convergence of BHI through Dagum Gini coefficient, kernel density estimation, Markov chain method and convergence test method. The study found that the overall gap of BHI showed an upward trend at the end of the investigation period compared with the beginning of the period, and the regional difference was the most important source of the significant difference in the development level of BHI in China.The internal differentiation trend of development level of BHI is expanding, and it is difficult to transfer upward within a certain threshold, but once it exceeds a certain threshold, it will have a greater probability to transfer to a higher level. The results of σ convergence, β convergence and club convergence indicate that there is absolute dispersion in the development level of BHI in the national and eastern regions.

Voronoi图是地学计算中的一个基本结构,但是在地形曲面上,它还缺乏能与平面Voronoi图媲美的精度和成熟的算法.在离散地形曲面的不规则三角网格网上引入计算几何的测地距离场,从格网边上的距离场奇点逐步生长代表平分线的双曲线,由双曲线的排列得到离散曲面的精确划分,再将划分的面片聚类,生成精确的测地Voronoi图(geodesic Voronoi diagram,GVD).然后,从定量与定性两方面对精确Voronoi图进行了检验,证明GVD可以给地形曲面空间分析带来基础性改进.基于奇点生长和双曲线排列的直接算法避免了现有算法对格网面片的过度细分与预处理,整体上直观易行,为数字地形分析发展严密的Voronoi图分析提供了有益探索.

With the acceleration of the urbanization process, the traditional urban management has become increasingly unable to meet the needs of urban management and development. At the same time, with the rapid development of artificial intelligence (AI) and big data (BD), the use of AI and BD to analyze cities has been gradually emerging. Therefore, this paper used AI and BD to study the optimization method of sustainable development of smart city public management. The research showed that the respondents in N, Z, and S cities were 60.67%, 60.07%, and 60.31% satisfied with the handling of events by urban public management subjects, respectively. The experts' evaluation scores on the feasibility and effectiveness of urban public management optimization strategies were 88.79 and 92.82, respectively. The public's satisfaction with the smart city public management subject's handling of events was still not high enough. The optimization strategy for sustainable development of smart city public management proposed in this paper with BD had certain practical value.