Hoang-Hon Trinh

Image-based Structural Analysis of Building using Line Segments and their Geometrical Vanishing Points

This paper describes an approach to detect and analyze the properties of building in image. We use line segments and belongings in the appearance of building as geometrical and physical properties respectively. The geometrical properties are represented as principal component parts (PCPs) as a set of door, window, wall and so on. As the physical properties, color, intensity, contrast and texture of regions are used. Analysis process is started by detecting straight line segments. We use MSAC to group such parallel line segments which have a common vanishing point. We calculate one dominant vanishing point for vertical direction and five dominant vanishing points in maximum for horizontal direction. A mesh of basic parallelograms is created by one of horizontal groups and vertical group. Each mesh represents one face of building. The PCPs are formed by merging neighborhood of basic parallelograms which have similar colors. The wall regions of PCPs are detected. Finally, the structure of building is described as a system of hierarchical features. The building is represented by number of faces. Each face is regarded by a color histogram vector. The color histogram vector just is computed by wall region of face. The proposed approach was used to recognize a database containing 1005 images and 115 queried images. It has been confirmed with various kinds of images taken for different conditions like camera systems, weather and seasons

Facet-based multiple building analysis for robot intelligence

This paper describes an approach to segment and recognize multiple buildings in the urban environment for robot intelligence. By grouping line segments which coincide with a common vanishing point, the non-building and building images are distinguished. The facets of building are detected and represented by the meshes of skewed parallelograms. The doors, wall region and windows are then estimated by merging the skewed parallelograms with similar color. To recognize a test image, each facet is described by its area, wall color histogram and a list of scale invariant feature transform (SIFT) descriptors. We selected a small number of SIFT features adapted with visual properties of buildings to represent the facet. To analyze multiple buildings, maximum numbers of dominant vanishing points are calculated for vertical and horizontal directions are one and five, respectively. In the first experiment, a set of 880 images is classified into building and non-building images. The second experiment is for recognizing a set of 80 test images from 500 image database. All images were taken from more than 100 buildings in Ulsan metropolitan city in South Korea under different conditions like viewpoints, camera systems, weather and seasons. We obtain 97% and 97.5% rate of correct segmentation and recognition, respectively.

Supervised training database by using SVD-based method for building recognition

This paper describes an approach to build a common model of building from different viewpoints. Then we apply to recognize building surfaces. For each image, buildingpsilas characters such as facets, areas, hue color histogram and a list of local features are calculated by our previous works. All correspondent facets are selected by supervision of user when the database is training. To calculate the characters of common model, we proposed a new method by using singular value decomposition (SVD). Given two or more similar vectors, SVD-based method computes an approximate vector which not only represents to the components but also automatically reduces the random noise. By using the common model, the number of facets and local features in the database are remarkably reduced. Therefore, the recognition rate is improved.

Supervised training database for building recognition by using cross ratio invariance and SVD-based method

This paper describes an approach to training a database of building images under the supervision of a user. Then it will be applied to recognize buildings in an urban scene. Given a set of training images, we first detect the building facets and calculate their properties such as area, wall color histogram and a list of local features. All facets of each building surface are used to construct a common model whose initial parameters are selected randomly from one of these facets. The common model is then updated step-by-step by spatial relationship of remaining facets and SVD-based (singular value decomposition) approximative vector. To verify the correspondence of image pairs, we proposed a new technique called cross ratio-based method which is more suitable for building surfaces than several previous approaches. Finally, the trained database is used to recognize a set of test images. The proposed method decreases the size of the database approximately 0.148 times, while automatically rejecting randomly repeated features from the scene and natural noise of local features. Furthermore, we show that the problem of multiple buildings was solved by separately analyzing each surface of a building.

OBJECTS SEGMENTATION USING MULTIPLE FEATURES FOR ROBOT NAVIGATION ON OUTDOOR ENVIRONMENT

This paper presents the method to recognize objects for autonomous robot navigation in outdoor environment. The proposition of the method segments from an image taken by a moving robot in an outdoor environment. The method begins with object segmentation, which uses multiple features to obtain the object of segmented region. Multiple features are color, context information, line segments, edge, Hue Co-occurrence Matrix (HCM), Principal Components (PCs) and Vanishing Points (VPs). We model the objects of outdoor environment that define their characteristics individually. We segment the region as a mixture using the proposed features and methods. Objects can be detected when we combine predefined multiple features. Next, the stage classifies the object into natural and artificial ones. We detect sky and trees of natural objects. And we detect building of artificial objects. The last stage shows the combination of appearance and context information. We implement the result of object segmentation using multiple features through experiments.

Urban building detection by visual and geometrical features

This paper describes an approach to detect the buildings in the urban environment. Visual and geometrical features of line segments are used to classify the building in the images. The buildings are also distinguished with other objects like sky, tree, bush and roads. Firstly, the line segments of building and non-building patterns are separated. The natural features are the contrast between two neighbored regions of segment, vanishing points, the appeared density, the vertical and horizontal alongside distributions. Those features are used to step-by-step reduce the segments of non-building pattern. The rests called the basic segments are grouped to create a mesh of skewed parallelograms. Each mesh represents a partial face of buildings. Finally, the faces or facets of building are detected by combining the neighbored partial faces. The building facet is refined again by its area. The proposed approach has been experimented for over 800 test images with the high rate of detection results.

Building Surface Refinement Using Cluster of Repeated Local Features by Cross Ratio

This paper describes an approach to recognize building surfaces. A building image is analyzed to extract the natural characters such as the surfaces and their areas, vanishing points, wall region and a list of SIFT feature vectors. These characters are organized as a hierarchical system of features to describe a model of building and then stored in a database. Given a new image, the characters are computed in the same form with in database. Then the new image is compared against the database to choose the best candidate. A cross ratio based algorithm, a novel approach, is used to verify the correct match. Finally, the correct match is used to update the model of building. The experiments show that the approach method clearly decreases the size of database, obtains high recognition rate. Furthermore, the problem of multiple buildings can be solved by separately analyzing each surface of building.

Window extraction using geometrical characteristics of building surface

This paper describes an approach to extract windows by analyzing geometrical characteristics of building surface. Firstly, building surfaces are detected and then wall region is extracted by using hue color of pixel; this step was well described in our previous works. The nonwall regions are considered as candidates of other components of building such as windows, doors, columns and so on. To extract the windows, the image of candidates is recovered in rectangular shape. Then the ambiguous candidates which have irregular shape, for example, long and thin or very small are coarsely rejected. The geometrical characteristics such as the center coordinates, area, aspect ratio and the aligned coexistence are used for extracting the windows. The proposed approach has been experimented for a database with 150 building surfaces comprising 1607 windows. We obtained 93.34% extraction rate.

Object recognition of outdoor environment by segmented regions for robot navigation

This paper describes a method to know objects in outdoor environment for autonomous robot navigation. The proposition of the method segments and recognizes the object from an image taken by moving robot in outdoor environment. Features are color, straight line, edge, HCM (Hue Co-occurrence Matrix), PCs (Principal Components), vanishing point and geometrical information. We classify the object natural and artificial. We detect tree of natural object and building of artificial object. Then we define their characteristics individually. In the process, we segment regions objects included by preprocessing. Objects can be recognized when we combine predefined multiple features. The correct object recognition of proposed system is over 92% among our test database which consist about 1200 images. We confirm the result of image segmentation using multiple features and object recognition through experiments.

Entrance detection of buildings using multiple cues

This paper describes an approach to detect the entrance of building with hopeful that it will be applied for autonomous navigation robot. The entrance is an important component which connects internal and external environments of building. We focus on the method of entrance detection using multiple cues. The information of entrance characteristics such as relative height and position on the building is considered. We adopt the probabilistic model for entrance detection by defining the likelihood of various features for entrance hypotheses. To do so we first detect buildings surfaces. Secondly, wall region and windows are extracted. The remained regions except the wall region and windows are considered as candidate of entrance. Finally, the entrance is identified by its probabilistic model.