Behnam Ghasemzade Qurmic; Alireza Safdarinejad
Abstract
Extended Abstract
Introduction
Analyzing the image blocks captured before and after geometrical changes is known as the conventional approach for detecting them in photogrammetric applications. Developed methods can be categorized into 1- comparison of 3D models generated via the image blocks and 2- ...
Read More
Extended Abstract
Introduction
Analyzing the image blocks captured before and after geometrical changes is known as the conventional approach for detecting them in photogrammetric applications. Developed methods can be categorized into 1- comparison of 3D models generated via the image blocks and 2- direct comparison of single images. The occurrence of radiometric differences in the geometrically changed areas can increase their discrimination and facilitate their detection. However, the occurrence of geometric changes without sensible radiometric effects is a special type of change that its identification is faced with more challenges. Slight displacement of the objects in the scene, small landslides, subsidence or uplift, the effects of local pressure and tension on objects in the industrial procedures and etc. are some examples of geometric changes that do not have a noticeable radiometric appearance in the images.
In the absence of incorrect observations, simultaneous triangulation of image blocks captured before and after geometric changes is a simple and effective way of reaching to detection of changes. In other words, by identifying the corresponding points in the fixed regions of the scene in the image blocks, the simultaneous triangulation of the image blocks captured in both epochs can align them in a unique object coordinate system. Thus, it can be possible to generate two independent and co-registered 3D models for identifying the occurred changes. However, maintaining the radiometric similarity of the changed areas leads to the identification of wrong-matched points when using automatic image matching methods.
The inclusion of an unknown 3D position for each wrong-matched point in the changed areas leads to a defect in the design of the mathematical model for the bundle adjustment. These defects result in incorrect generation of the 3D models, large and systematic errors in the residuals of observations, and incorrect estimation of the extrinsic parameters of images. The remedy to this defect is to assign two distinct unknown 3D positions for each wrong-matched point before and after changes in the bundle adjustment. Lack of prior knowledge of the wrong-matched points located in the changed areas is the cause of this problem. In this article, an iterative solution is proposed to identify and correct the effects of the wrong-matched points in the process of simultaneous bundle adjustment.
Materials and Methods
In the proposed method, at first, all the confident radiometrically matched points among all images taken before and after the geometric changes are detected via the well-known feature-based image matching methods. Their matched positions, then, are again accurately rectified and verified by the least squares image matching method. The matched points identified after refinement are classified into two categories. 1- The matched points that have been detected only in the images of one image block and 2- The matched points that have been detected at least in two images in each image block. Among the points of the second category, there probably are matched points that are geometrically changed between two epochs, but their radiometric similarities have made to incorrectly identified as the matched points between two image blocks. In this paper, these were called the wrong-matched points which are iteratively identified and their corresponding mathematical models are corrected in the triangulation process.
To do so, three different bundle adjustments are performed as the first step. Independent triangulation of the image blocks captured before and after the geometric changes and the simultaneous bundle adjustment of both blocks via the initially detected matched points of the first and second categories are the first three triangulations. Due to the existence of wrong-matched points, the initial simultaneous triangulation has a defect in the design of the mathematical model, which is gradually and in an iterative process, the wrong-matched points located in the changed areas would be identified.
Identification of the wrong-matched points is done using the relative comparisons on their residual vectors. The comparisons are designed in two consecutive statistical tests. The main idea of this method has been inspired by the well-known Baarda test in the detection of gross errors in the observations of geodetic networks. By gradual identification of the wrong-matched points, their corresponding mathematical model will be modified in the bundle adjustment.To do so, the unknown values of the 3D coordinates of these points are separated for the time before and after the change epochs.This action by modification of the mathematical model in the bundle adjustments brings back the relative equilibrium in the estimation of the residual vector of observations.
Results and Discussion
Implementation and comparison of the proposed method with a conventional geometric approach in identifying the incorrectly matched points (using robust estimation of the epipolar geometry) have shown the adequacy and superiority of the proposed method. The proposed method, on average in more than 11 different experiments, was able to achieve an average accuracy of 85.8% in identifying the changed points. Meanwhile, the proposed method shows a 34.5% improvement compared to the conventional geometric approach based on epipolar geometry.
Conclusions and suggestions
The proposed method is an effective solution for identifying the geometrically changed points in the simultaneous triangulation of image blocks before and after geometric changes when the changed areas have a stable radiometric similarity. This method is more sensitive to the occurred changes than the conventional method of identifying incorrect correspondences based on epipolar geometry. Iterative adjustment of the observations’weight matrix through the Variance Components Estimation (VCE) techniques in order to detect and eliminate the effects of wrong-matched points can be considered a future research topic in this field.
Alireza Taheri Dehkordi; Seyyed Mohammad Milad Shahabi; Mohammad Javad Valadan Zouj; Mahmood Reza Sahebi; Alireza Safdarinejad
Abstract
Extended Abstract
Introduction
Over the past three decades, with the rapid development of spatial-based satellite imagery, remote sensing technology has found a special place in various applications of urban management. Production of status maps of urban structures, the study of energy loss status, ...
Read More
Extended Abstract
Introduction
Over the past three decades, with the rapid development of spatial-based satellite imagery, remote sensing technology has found a special place in various applications of urban management. Production of status maps of urban structures, the study of energy loss status, identification of thermal islands, monitoring of urban vegetation, and assessment of air pollution are just a few examples of areas related to urban management that remote sensing technology is the basis for indirect measurement of the related quantities. Maps of urban structures such as building blocks are commonly used in crisis management, urban design, and urban development studies.
Materials
In this study, the production of urban building block maps using Sentinel 1 and 2 satellite images has been conducted. Normalized Difference Vegetation Index (NDVI) and Normalized Difference Building Index ( NDBI ) for three consecutive months, the slope feature derived from the 30-meter Shuttle Radar Topographic Mission (SRTM)Digital Elevation Model of the study area, along with two Vertical – Vertical (VV) and Vertical - Horizontal ( VH ) polarization in both ascending and descending orbits, form the set of input features.
Methods
The proposed method of this paper relies on the use of a generalizable trained classifier. Initially, the classifier is trained in 2015 using training samples obtained from a new rigorous refining process using different remote sensing and spatial products. This rigorous refining process uses a reference urban map of 2015. In the first step, the corresponding areas related to the ways and roads are removed using the OpenStreetMap data layer. Areas suspected of vegetation with NDVI greater than 0.2 are then discarded. Also, due to the high backscattering of buildings in Synthetic Aperture Radar images, areas with a value less than the average backscattering coefficient of the remaining areas are eliminated. Finally, the residual map is refined using the Mahalanabis distance and the Otsu automatic thresholding method. The trained classifier is then used to generate a map of building blocks at similar time intervals for the three target years (2018, 2019, and 2020). Due to the diversity of texture and density of building blocks in the metropolis of Tehran, the proposed method has been evaluated in this area. Due to the concentration of political, welfare, and social facilities, Tehran has experienced more unplanned and irregular expansion and urbanization than other cities in Iran, which has lead to changes in buildings and constructions. Also, due to the availability of free satellite images and various online processing operations, the Google Earth Engine platform has been used in this study. The performance of three different classifiers including Random Forest (RF), Minimum Mahalanabis Distance (MD), and Support Vector Machines (SVM) are examined in this process. In order to evaluate the proposed method, reference samples obtained from visual interpretation of high-resolution satellite images (Google Earth) in all three target years have been used.
Results
The performance of the aforementioned classifiers has been investigated using 3 different criteria: overall accuracy, user accuracy, and F-score of building blocks. The RF method with an overall accuracy of over 93% in all three target years has shown the best performance. The SVM method ranks second with an accuracy of about 91% every three years. However, the MD method with an overall accuracy below 85% in all three target years has not performed well.
Discussion
The results show better performance of the RF method in all three target years with an overall accuracy of over 93%. It should be noted that the MD classifier with higher user accuracy than other methods, has shown better performance in detecting the class of building blocks. However, the RF method is the best classifier in terms of the user accuracy of the background class. The effect of using two VV and VH polarization and also the slope derived from the SRTM Model in the input feature set on the final accuracy of classification was also investigated. According to the results, the simultaneous use of these three features produces more accurate results in both target classes. However, the results show that the use of VV polarization increases the final classification accuracy compared to VH polarization. The presence of slope feature along with both polarizations has also increased the classification accuracy of each class, especially the background class. However, the exclusion of both VV and VH features from the input feature set has resulted in a more than 10% reduction in overall classification accuracy.
Conclusion
Based on calculated overall accuracies which are above 80% in the majority of investigated cases, two different results can be concluded. First, the trained classifier has shown good temporal generalization and has achieved acceptable accuracy in the target years. Second, due to the different collection processes of training and evaluation data, the proposed rigorous refining method for the preparation of training data has shown good performance. The effect of using two VV and VH polarization and also the slope derived from the SRTM Digital Elevation Model in the input feature set on the final accuracy of classification was also investigated. According to the results, the simultaneous use of these three features produces more accurate results in both target classes. However, the results show that the use of VV polarization increases the final classification accuracy compared to VH polarization. The presence of slope feature along with both polarizations has also increased the classification accuracy of each class, especially the background class. However, the exclusion of both VV and VH features from the input feature set has resulted in a tangible decreasein overall classification accuracy.