ArcGIS Desktop逆引きガイド : バージョン10対応

The world is covered with millions of buildings, and precisely knowing each instance's position and extents is vital to a multitude of applications. Recently, automated building footprint segmentation models have shown superior detection accuracy thanks to the usage of Convolutional Neural Networks (CNN). However, even the latest evolutions struggle to precisely delineating borders, which often leads to geometric distortions and inadvertent fusion of adjacent building instances. We propose to overcome this issue by exploiting the distinct geometric properties of buildings. To this end, we present Deep Structured Active Contours (DSAC), a novel framework that integrates priors and constraints into the segmentation process, such as continuous boundaries, smooth edges, and sharp corners. To do so, DSAC employs Active Contour Models (ACM), a family of constraint- and prior-based polygonal models. We learn ACM parameterizations per instance using a CNN, and show how to incorporate all components in a structured output model, making DSAC trainable end-to-end. We evaluate DSAC on three challenging building instance segmentation datasets, where it compares favorably against state-of-the-art. Code will be made available.

Learning deep structured active contours end-to-end

Using very high resolution remote sensing images to extracting urban features from very high resolution remote sensing images is a very complex and difficult task. The improvement in geospatial technologies brought forward many solutions that can help in improving the process of urban feature extraction. Data collection using light detection and ranging (LiDAR) and capturing very high resolution optical images concurrently is one of these solutions. This research proves that the fusion of high-resolution optical image with LiDAR data can improve image processing results. It is based on increasing urban features extraction success rate by reducing oversegmentation. The fusion process relies first on wavelet transform techniques, which are run several times with different parameters (rules). Then, an innovative technique is implemented to improve fusion process. The two techniques are compared, and both have reduced fragmented segments and created homogeneous urban features. However, the fused image with the innovative technique has improved the accuracy of the segmentation results. The average accuracy for building detection is 96% (maximum 100% and minimum 92%) using the innovative technique compared to 21% and 51% for no fusion and wavelet-fusion-based techniques. Furthermore, an index is used to measure the quality of the building details which are detected after using the innovative fusion technique. The result indicates that the quality index is greater or equal to 86%.

Toward Robust Segmentation Results Based on Fusion Methods for Very High Resolution Optical Image and LiDAR Data

There is an increase in the collection and availability of LiDAR data across Canada. While digital terrain and surface models are the most common derivative products, the LiDAR point cloud data can also be used to extract various layers of information, including vegetation, utility lines, bridges, and buildings. This paper describes the current initiative by the Canadian federal government to derive building footprints from LiDAR data in order to generate a building footprint dataset for Canada’s Open Data portal and to evaluate the minimum acceptable criteria for successful and accurate building extraction. The results provide guidelines for the minimum density requirements of LiDAR point cloud data for the extraction of building footprints in an urban setting. Two criteria were tested: (i) minimum point density and (ii) the effect of vendor-classified vs re-classified (classified using open-source tools) point data clouds in the building extraction process. Results indicate that vendor-classified point cloud data with a minimum density of 4 pts/m2 is sufficient to accurately extract building footprints with >75% confidence. For re-classified LiDAR a density of at least 8 pts/m2 would be required to meet this confidence level. However, commission errors found in the fully automatic re-classified method are numerous and manual editing or further algorithm refinements are necessary before it could be used in production.

Extraction of Building Footprints from LiDAR: An Assessment of Classification and Point Density Requirements

The classic active contour model raises a great promising solution to polygon-based object extraction with the progress of deep learning recently. Inspired by the physical vibration theory, we propose a contour vibration network (CVNet) for automatic building boundary delineation. Different from the previous contour models, the CVNet originally roots in the force and motion principle of contour string. Through the infinitesimal analysis and Newton's second law, we derive the spatial-temporal contour vibration model of object shapes, which is mathematically reduced to second-order differential equation. To concretize the dynamic model, we transform the vibration model into the space of image features, and reparameterize the equation coefficients as the learnable state from feature domain. The contour changes are finally evolved in a progressive mode through the computation of contour vibration equation. Both the polygon contour evolution and the model optimization are modulated to form a close-looping end-to-end network. Comprehensive experiments on three datasets demonstrate the effectiveness and superiority of our CVNet over other baselines and state-of-the-art methods for the polygon-based building extraction. The code is available at https://github.com/xzq-njust/CVNet. 

CVNet: Contour Vibration Network for Building Extraction

Here we describe and apply a semi-automated, object-based method for extracting vector-building footprint polygons from aerial photographs (orthophotos) within urban settings. The approach integrates the use of high resolution orthophotos and image segmentation software and is compared with methods using Light Detection and Ranging (LiDAR) as the source data input. LiDAR data gives the best results with less processing, but is not widely used by municipalities due to the expense. Results from semi-automated image segmentation of the orthophotos showed a high accuracy between extracted building segments and reference building footprints for two study sites, comparable to those achieved using LiDAR data. We recommend image acquisition during summer months with a resolution of 10 cm by 10 cm. When data acquisition budgets are limited, combining ancillary GIS on roads with a semi-automated and object-based segmentation approach is a best practice strategy for land cover feature extraction and change quantific...

/pdf/semi-automated-building-footprint-extraction-from-2f8e3zkjkf.pdf

Semi-Automated Building Footprint Extraction From Orthophotos

Spatial analysis methods used for detecting, interpolating, or predicting local patterns in geographic data require delineating a neighbourhood to define the extent of the spatial interaction. Certain spatial analysis methods, such as interpolation, have implemented the concept of directionality and barriers. However, not all approaches take into consideration geographic or environmental constraints such as impassable mountain ranges, road networks, or coastlines. Specifically, complex marine landscapes and coastlines pose problematic neighbourhood definitions for standard neighbourhood matrices used in the spatial analysis of marine environments. Here, we offer a new approach to constraining spatial neighbourhoods when conducting geographical analysis in marine environments. We developed methods and open source software (COnstraining SpatIal NEighbourhoods—COSINE) for modifying spatial neighbourhoods, and demonstrate their utility with a marine study of oil spills. The COSINE graphical user interface all...

NelsonTrisalyn

Papers

Semi-Automated Building Footprint Extraction From Orthophotos

Cosine: A Tool for Constraining Spatial Neighbourhoods in Marine Environments