Label placement is a critical and challenging aspect of automated cartography. Traditional methods of automated map label placement mainly focus on vector maps or digital line graphics, and these methods often perform poorly when handling label placement on remote sensing images. This study proposes a method for building label placement that considers the visual characteristics of the image. This method first determines candidate labeling areas through superpixel segmentation and k-means clustering. Then a series of candidate labeling lines are derived using morphological computation, graph theory and other techniques. The labeling line that best fits the structural characteristics of the building is selected according to a predefined optimization strategy. Finally, the rotation angle of each character in the label is calculated according to the labeling line of the building, and the appropriate labeling style is determined based on the brightness characteristics of the labeling area. Compared with traditional methods, the proposed method thoroughly analyzes the visual features of building images, such as structure, color, texture, and brightness. During the label placement process, the proposed method comprehensively considers the texture complexity and morphological information of the label placement area, providing a more readable, natural, and aesthetically pleasing label placement solution.

With the advancement of vision-based autonomous driving technology, pedestrian detection have become an important component for improving traffic safety and driving system robustness. Nevertheless, in complex traffic scenarios, conventional pose estimation approaches frequently fail to accurately reconstruct occluded keypoints, primarily due to obstructions caused by vehicles, vegetation, or architectural elements. To address this issue, we propose a novel real-time occluded pedestrian pose completion framework termed Separation and Dimensionality Reduction-based Generative Adversarial Imputation Nets (SDR-GAIN). Unlike previous approaches that train visual models to distinguish occlusion patterns, SDR-GAIN aims to learn human pose directly from the numerical distribution of keypoint coordinates and interpolate missing positions. It employs a self-supervised adversarial learning paradigm to train lightweight generators with residual structures for the imputation of missing pose keypoints. Additionally, it integrates multiple pose standardization techniques to alleviate the difficulty of the learning process. Experiments conducted on the COCO and JAAD datasets demonstrate that SDR-GAIN surpasses conventional machine learning and Transformer-based missing data interpolation algorithms in accurately recovering occluded pedestrian keypoints, while simultaneously achieving microsecond-level real-time inference.

Research on map generalization has led to many algorithms for multiple elementary processes, such as object selection, aggregation, simplification, and displacement. Algorithms for different processes are usually combined to workflows or orchestrated using multi-agent systems. Here, we present a novel approach integrating object selection and displacement at an algorithmic level. We model both processes together as an optimization problem in the form of a mixed-integer quadratic program and demonstrate that it can be optimally solved using a mathematical problem solver. Moreover, we present an efficient heuristic. In experiments with roads and buildings from OpenStreetMap, our methods showed a good capability to unselect a small set of buildings whose inclusion in the output map would have caused large displacements or proximity conflicts. For a quantitative evaluation, we solved a benchmark instance once with our new model integrating selection and displacement and once with a variant of our model where the selection of objects was prescribed based on a solution found with an existing approach via simulated annealing. Comparing the two models, our integrated model yielded a solution of 33% less total cost. We conclude the article with a discussion of possible follow-up work.

Computer screens often constrain the level of detail and clarity of displays. High-density data require a predefined strategy to select significant features hierarchically to allow interactive data zooming. Although many methods are available for hierarchically selecting rivers from vector data, some approaches for raster data are better than others for maintaining accuracy when the original river data are in a raster format during generalization. In this study, a raster-based approach is proposed to allow hierarchical superpixel selection in river networks. Linear spectral clustering segmentation was applied to divide the original raster river networks into superpixels at multiple levels. A graph was constructed to organize the generated river network superpixels based on the distances between adjacent superpixels by considering the weights determined by the four types of rules. Finally, the total weight values were ranked, the river-network superpixels were selected according to their weights, and the redundant pixels at the river-network intersections were removed. Compared with the traditional vector selection method, the proposed superpixel river network selection method can effectively consider the characteristics of river width without artificial river grading and preserve the main structure and connectivity features during hierarchical mapping. Notably, the average geometry and density changes decreased by 15.8% and 5.1%, respectively.

Numerous methods based on square rasters have been proposed for polygon generalization. However, these methods ignore the inconsistent distance measurement among neighborhoods of squares, which may result in an imbalanced generalization in different directions. As an alternative raster, a hexagon has consistent connectivity and isotropic neighborhoods. This study proposed a hexagon-based method for polygon generalization using morphological operators. First, we defined three generalization operators: aggregation, elimination, and line simplification, based on hexagonal morphological operations. We then used corrective operations with selection, skeleton, and exaggeration to detect, classify, and correct the unreasonably reduced narrow parts of the polygons. To assess the effectiveness of the proposed method, we conducted experiments comparing the hexagonal raster to square raster and vector data. Unlike vector-based methods in which various algorithms simplified either areal objects or exterior boundaries, the hexagon-based method performed both simplifications simultaneously. Compared to the square-based method, the results of the hexagon-based method were more balanced in all neighborhood directions, matched better with the original polygons, and had smoother simplified boundaries. Moreover, it performed with shorter running time than the square-based method, where the minimal time difference was less than 1 min, and the maximal time difference reached more than 50 mins.

Remote sensing mapping plays an important role in understanding regional development and geographical environment characteristics. Traditional remote sensing mapping at different levels usually fails to consider the shape, quantity, distribution, and position features of map objects. Therefore, a multilevel representation of urban buildings is realized based on the proposed framework for multilevel mapping from remote sensing images. In this process, the Mask R-CNN method is first applied to extract buildings from remote sensing images. Then, the orthogonal shape features of the extracted buildings are reconstructed based on corner detection, and urban roads are generated by extracting the internal structural characteristics of urban buildings for further multilevel representation. Finally, three innovative raster-based generalization algorithms, including simplification, aggregation, and typification based on Hough line detection technology, are developed for a multilevel representation of urban buildings. The experimental results reveal that the proposed methods can effectively realize multilevel mapping of urban buildings from remote sensing images while meeting basic cartographic requirements.

Land cover multiresolution mapping of remote sensing images contributes greatly to land-use management, environmental protection, and city planning. In traditional mapping of this type, the representation of different land-use types depends on the image resolution, and the geometric, topologic, and semantic characteristics are not considered. This approach can cause a loss of useful information and the redundancy of useless information. In this study, we propose a superpixel-based land cover (multiresolution representation SULR) method for remote sensing images that employs multifeature fusion. In this process, we first define three basic superpixel operations, collapse, connection, and cutting, as the basic operators of multiresolution land cover mapping. Then, the topological adjacent land parcels are combined through the amalgamation of polygons with heterogeneous properties and aggregation of polygons with homogeneous properties based on the three proposed superpixel operators. Finally, the geometric boundaries of parcels are simplified by combining the superpixel collapse operator and image thinning technologies. Compared with traditional image scale transformation methods, the proposed method can more effectively achieve multiresolution mapping of land cover from remote sensing images by considering the geometric, topologic, and semantic characteristics of land parcels.

One important classical research area in automated cartographic generalization is simplification. Over the past few decades, numerous scholars have proposed various methods for polygon and line simplification, most of which have focused on vector data. However, with the rapid development of computer vision technology, unstructured image analysis and processing has provided a plethora of information, as well as new challenges. Therefore, in this article, we propose a new method for simplifying polygonal and linear features: a superpixel segmentation (SUSS) method specially designed for image data. In this method, polygonal boundaries are first divided by a superpixel algorithm called simple linear iterative clustering. Then, three types of curves – convex, concave, and flat – are globally simplified by comparing and selecting superpixels. Finally, uneven local features are removed by Fourier descriptors. In addition, the proposed SUSS method is extended for linear features, and it maintains topological relationships. To demonstrate the effectiveness of this approach, we use contours and water area data to perform experiments. Compared with the classic Douglas–Peucker and Wang and Muller algorithms, the proposed method is able to properly simplify the curves of polygonal and linear features while maintaining their essential shapes, and it maintains a steady change in area for large-scale applications while effectively avoiding self-intersection issues. Compared with the typical smoothing and Raposo algorithms, the proposed SUSS method can simplify lines at different scales and guarantee effective smoothing while maintaining displacement.