Simplified Representation Of Map Elements From Computer Vision Perspective

With the rapid progress and development of human society,map data are also faced with problems such as great capacity of data,fast update and various forms of data,which leads to the continuous increase of labor cost for cartography and makes the existing cartography model unable to meet the growing demands for cartography.Therefore,it is an inevitable trend for the intelligent development of modern cartography to introduce robots into the field of cartography instead of manual operations.Computer vision is a subject that studies how to make the computer "see" the world.Computer vision is widely used,such as face recognition,unmanned driving,unmanned aerial vehicle flight and so on.There are some basic computer vision technologies such as image segmentation,image feature detection,image mosaic,target matching and recognition,and deep learning.Under the background of artificial intelligence technology,in this research,we use image data as the basic research object and introduce the relevant computer vision technologies for map generalization.For example,image corner detection technology is used for identification of map elements,superpixel segmentation is applied for simplified representation of geographical elements,image Fourier transform is used for reconstruction of geographic features.Combing the cutting-edge computer vision technologies and theories,we propose a series of new methods,which called 8S(SUSS-L,SUSS-A,SBIP,SUBS,SUPA-G,SUPA-B,SUCE,SURC),for simplification,aggregation and collapse of linear and polygonal features based on raster map,which mainly include the following four aspects:(1)Simplification of general linear features(such as contour lines)based on superpixel structure.One important classical research area in automated cartographic generalization is line simplification.In this article,we propose a new method for simplifying linear features: a superpixel segmentation(SUSS-L)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.To demonstrate the effectiveness of this approach,we use contours 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-L method can simplify lines at different scales and guarantee effective smoothing while maintaining displacement.(2)Simplification of general polygonal features considering geometric,geographical and visual features.For general polygonal features,such as lakes,we present a new line simplification approach based on image processing(SBIP),which is specifically designed for raster data.First,the key corner points on a multi-scale image feature are detected and treated as candidate points.Then,to capture the essence of the shape within a given boundary using the fewest possible segments,the minimum-perimeter polygon is calculated and the points of the minimum-perimeter polygon are defined as the approximate feature points.Finally,the points after simplification are selected from the candidate points by comparing the distances between the candidate points and the approximate feature points.By defining the visibility constraint of geographical features,this method is especially suitable for simplifying water areas as it is aligned with people's visual habits.In addition,the proposed SUSS-L method is extended for areal features,which called SUSS-A method in this parper,and it maintains topological relationships and considers the geographical characteristics of complex lakes.For manual polygonal features,such as buildings,we propose a new algorithm called superpixel building simplification(SUBS),based on image data.In this method,the buildings are first divided into two types by corner detection: buildings with orthogonal features and buildings with non-orthogonal features.Then,the buildings are globally simplified using a superpixel segmentation algorithm for superpixel extraction via energy-driven sampling.Finally,the buildings are locally simplified to preserve their geometric features.Compared with traditional algorithms,the results indicate that the proposed method can produce satisfactory results for the simplification of buildings with both orthogonal and non-orthogonal features and effectively preserve the area and centre of mass of the buildings.In addition,the SUBS method can generate different representation styles of buildings while effectively avoiding self-intersection.(3)Adaptive aggregation method of general polygonal features(such as lakes)and manual polygonal features(such as buildings).By corner detection: the polygonal features can be divided into polygons with orthogonal features and polygons with non-orthogonal features.For non-orthogonal polygonal features,such as lakes,the methodologies for aggregation of general polygons,which called SUPA-G,can be divided into three steps.First,superpixel segmentation of polygon groups using a superpixel algorithm called simple linear iterative clustering.Then,global aggregation based on superpixel selection.Finally,local adjustment of the aggregate boundaries using Fourier descriptors.The proposed method can be effectively used for the aggregation of general polygons such as water areas at different levels of detail.By using Fourier descriptors,the proposed method can produce smooth boundaries while maintaining global features.Similarly,the process of aggregation of buildings with orthogonal features,which called SUPA-B,can also be divided into three steps.First,superpixel segmentation of building groups using a superpixel algorithm called clustering extracted via energy-driven sampling.Then,global aggregation based on superpixel selection.Finally,local adjustment of the aggregate boundaries based on superpixel filling and removal.By integrating the superpixel method,the proposed method can be effectively used for the aggregation of buildings at different levels of detail.In addition,this method can preserve orthogonal features of the buildings while effectively avoiding self-intersection.(4)Progressive collapse of polygonal features.A new algorithm called superpixel centerline extraction(SUCE)for dual-line roads based on the raster data is proposed.In this method,dual-line roads are first segmented using a superpixel algorithm called simple linear iterative clustering.Then,the superpixels located at road intersections are merged to generate connection points from their skeletons.Finally,the centerlines of roads are generated by connecting the center points and edge midpoints of each superpixel.The results indicate that the proposed SUCE method can effectively extract centerlines from dual-line roads and restore the original road intersections while avoiding burrs and noises,both for simple and complex road intersections.A new algorithm called superpixel river collapse(SURC)is proposed to convert dual-line rivers to single-line rivers based on raster data.In this method,dual-line rivers are first segmented at different levels of detail using a superpixel method called simple linear iterative clustering.Then,by connecting the edge midpoints and center of mass of each superpixel,the single-line rivers are preliminarily generated from dual-line rivers.Finally,an interpolation algorithm called polynomial approximation with exponential kernel is applied to remain the uniform distribution of feature points of single-line rivers at different levels of detail.To test the SURC method,dual-line rivers from Guangdong province,China,are applied to perform experiment.Compared with three typical thinning algorithms,the results show that the SURC method can progressively generate smooth single-line rivers from dual-line rivers considering different river widths while effectively avoiding burrs.The algorithms based on computer vision technology are rich and mature,but few of them are applied for map generalization.In this study,relevant computer vision technologies are introduced for feature extraction,recognition and scale transform of map elements,and the corresponding geographical features are taken into account.Finally,the simplification,aggregation and collapse algorithms in map generalization are realized.We try to make a little contribution to the intelligent development of cartography in the era of artificial intelligence.