Knowledge Based 3d Building Model Recognition Using Convolutional Neural Networks From Lidar and Aerial Imageries

Abstract

ABSTRACT:In recent years, with the development of the high resolution data acquisition technologies, many different approaches and algorithms have been presented to extract the accurate and timely updated 3D models of buildings as a key element of city structures for numerous applications in urban mapping. In this paper, a novel and model-based approach is proposed for automatic recognition of buildings' roof models such as flat, gable, hip, and pyramid hip roof models based on deep structures for hierarchical learning of features that are extracted from both LiDAR and aerial ortho-photos. The main steps of this approach include building segmentation, feature extraction and learning, and finally building roof labeling in a supervised pre-trained Convolutional Neural Network (CNN) framework to have an automatic recognition system for various types of buildings over an urban area. In this framework, the height information provides invariant geometric features for convolutional neural network to localize the boundary of each individual roofs. CNN is a kind of feed-forward neural network with the multilayer perceptron concept which consists of a number of convolutional and subsampling layers in an adaptable structure and it is widely used in pattern recognition and object detection application. Since the training dataset is a small library of labeled models for different shapes of roofs, the computation time of learning can be decreased significantly using the pre-trained models. The experimental results highlight the effectiveness of the deep learning approach to detect and extract the pattern of buildings' roofs automatically considering the complementary nature of height and RGB information.

Multi-Task cGAN for Simultaneous Spaceborne DSM Refinement and Roof-Type Classification

“…Innovative architectures, such as CNNs, have demonstrated the ability to classify high dimensional data sources accurately and robustly and have become the state-of-the-art for image recognition tasks. Alidoost and Arefi [22] proposed an approach based on fine-tuning of a pre-trained CNN model, which accomplishes building segmentation, feature extraction, and building roof labeling to create an automatic recognition system for various types of buildings. The training is done on both spectral and depth images separately, and the final predicted roof shape is simply taken as the highest probability result between the two models.…”

Automatic 3-D Building Model Reconstruction from Very High Resolution Stereo Satellite Imagery

“…Each patch is assigned to a label manually based on its corresponding roof type [20]. The main difference between this approach and previous image patch dataset generation approaches [67,68] is that the main orientation of each roof is also considered inside the patch. Therefore, the quality of roof patches cannot be degraded by rotation and resizing.…”

Automatic Classification of Roof Shapes for Multicopter Emergency Landing Site Selection

“…The option of freezing initial layers is exploited with a variable number of frozen layers chosen. When layer 11 is said to be frozen, this means all previous layers, (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11), are frozen during training. All base architectures and tested hyper-parameters are shown in Table 1.…”

“…The GIS community has begun to apply CNNs to roof identification. Perhaps most closely related to this paper, Alidoost and Arefi (2016) trained CNNs using satellite (RGB) and digital surface map (DSM) images to label basic roof shapes [5]. However the final predicted roof shape was simply taken as the highest probability result between the two models.…”