Data-Driven Diffraction Loss Estimation for Future Intelligent Transportation Systems in 6G Networks

Question

1. How can UAVs and eVTOLs improve communication in 6G networks?

2. What is the purpose of DLIPHE algorithm?

3. What factors affect diffraction loss estimation?

4. How do methods based on high-resolution optical imagery estimate building heights?

Accepted Answer

UAVs and eVTOLs, termed as urban aerial devices (UADs), can improve communication in 6G networks by providing line of sight (LoS) to signals originating from consumer electronic devices. These UADs fly around 120 meters and use extremely high-frequency bands (EHF), which face challenges such as signal absorption by buildings, trees, and clouds. To address these limitations, UADs can utilize integrated satellite, aerial, and terrestrial networks. By understanding diffraction losses in outdoor environments, operators can decide the optimal communication path between a transceiver and receiver. Diffraction losses occur when UADs fly over buildings and close to rooftops. The height of buildings and tall obstacles is a key parameter to estimate diffraction losses. Previous technologies for determining building heights include high-spectral satellite optical image data, LiDAR data, and synthetic aperture radar (SAR) image data. However, these optical approaches are costly and difficult to implement on a wide scale. Street-view imagery has been used for estimating building heights, but it may not provide accurate estimations for UADs. Open-source mapping platforms like Google Street View, Streetside, and OpenStreetMap offer opportunities for data collection, but they may not provide real-time and automatic building height estimations for UADs. Therefore, understanding diffraction losses and estimating building heights are crucial for improving communication in 6G networks using UAVs and eVTOLs.

Accepted Answer

The DLIPHE algorithm aims to estimate a building's height using publicly available, static, non-interactive Google Street View images. It utilizes semantic segmentation to identify buildings and detect their height by analyzing the contour of the images. This advanced image processing technique allows for accurate real-time height prediction with minimal complexity, enabling efficient communication path planning. The algorithm creates an improved end-to-end real-time system that can determine the heights of all buildings in the path of Unmanned Aerial Devices (UADs) using online 2D image data and footprints.

Accepted Answer

Diffraction loss estimation depends on the height of obstacles and the distance between communication ends. ITU's mathematical models provide detailed insights into this relationship. Studies like [23] and [26] have proposed models for specific scenarios, such as rainforest environments and thin isolated trees. However, these models require manual inputs for accurate predictions. Statistical models, while useful, cannot provide real-time diffraction loss information to UADs. Therefore, researchers are exploring building height estimation methods to optimize propagation paths for UADs. These methods aim to enable automatic adjustments in communication paths, enhancing the efficiency of UADs in various environments.

Accepted Answer

Methods based on high-resolution optical imagery estimate building heights using shadows cast and acquisition geometry. Studies utilize high-resolution satellite images to analyze shadows and geometry for height estimation. Additionally, aerial LiDAR data constructs polyhedral building roof-tops based on structure or shape. A convolutional-deconvolutional DNN framework proposed in [30] maps single satellite imagery to a digital surface model for urban area analysis. This framework combines optical imagery data and LiDAR data as training data, improving alignment through calibration processes. The model's validation on a high-resolution dataset of central Dublin demonstrates the reconstruction of a 3D model from single-view aerial imagery. However, these methods face scalability challenges due to high costs, extensive data collection, and computational resource requirements, limiting real-time planning capabilities for future extreme situations. Therefore, focusing on real-time solutions like image processing is crucial.

Accepted Answer

Building height estimation using street-view data involves several steps. First, building rooflines and corners are extracted from Google Street View API, based on their parameters and 3D maps footprint. Then, building corners are used instead of corner lines, yielding better results in camera location calibration. A deep neural network-based method is applied to filter and segregate the roofline and the building's corner. Finally, the height of the building is calculated via the pinhole camera model, based on the building's roofline and valid corners from the DNN model. This method is highly scalable and efficient, as street-scene data are easily available from Microsoft Streetside and Google Street View API. Other methods, such as those proposed by Zhao et al. [18], Yuan et al. [31], Mou et al. [32], and Diaz et al. [16], also utilize street-view imagery for building height estimation, but with different approaches and techniques. Overall, street-view data provide a valuable resource for building height estimation due to their high-quality, detailed imagery and ease of accessibility.

Accepted Answer

UADs transmit data in urban environments by using multiple input and output antennas for transmission. They are deployed in urban environments with several buildings, flying over 120 meters and more. UADs transmit data on the 28 GHz spectrum for faster data rates. They start from an initial point and choose the shortest trajectory path to a fixed destination point. The focus is on a UAD serving as a transmitter and a receiver (RX) for analyzing diffraction loss caused by buildings. Rooftops of buildings are assumed to be flat, allowing for single knife-edge diffraction. The Fresnel diffraction model is used to express signal losses, considering factors like height distance, distance between devices, and wavelength. The ray diagram in Figure 2 illustrates this model. UADs also estimate building heights around them using DLIPHE method for better communication.

Accepted Answer

The DLIPHE model follows a series of steps to estimate building heights. Initially, Google Street View API is used to download static images of the building of interest. Metadata, including camera location, is obtained alongside the imagery. Semantic segmentation identifies the building using transfer learning. Advanced image processing techniques extract building heights. OSM API is then used to extract building footprint data. Finally, computations are performed to estimate the building's height. Algorithm 1 provides detailed instructions for this process, as illustrated in Figure 3.

Accepted Answer

For Google Street View API request, parameters such as location, size, heading, and pitch are required. These parameters help in obtaining the desired building's image in a specific manner. The location parameter specifies the geographical coordinates of the building, while the size parameter determines the resolution of the image. The heading and pitch parameters control the angle and elevation of the image capture, allowing for different perspectives and distances from the building. In urban areas, a closer distance with higher elevation is often needed due to the density of buildings, which may result in a slightly slanted appearance. Additionally, the API allows for downloading a metadata file containing the camera's location, which can be used to calculate the distance between the building and the camera.

Accepted Answer

Image segmentation is a digital image processing technique used for partitioning an image into multiple segments or regions. It locates object boundaries in images and simplifies further analysis by grouping pixels based on specific characteristics like intensity, color, or connectivity. Semantic segmentation, a pixel-level prediction, categorizes each pixel in an image according to a group or class, linking pixels to specific class labels. It differs from object detection by not predicting bounding boxes or classifying different instances of the same object. Common applications include self-driving vehicles, human-computer interaction, and photo editing. Deep convolutional neural networks (DCNNs) have achieved remarkable results in semantic segmentation, with architectures like DeepLabv3+ and Xception being widely used.

Accepted Answer

Image thresholding is a digital image processing technique used for segmenting or classifying pixel values in an image. It assigns a pixel as black if its intensity is less than a predetermined constant T, and as white otherwise. The threshold value T is selected based on the object's intensity, size, number of objects, and image occupancy. Automatic thresholding may not yield desired results, so Otsu's thresholding method is used to find an optimal threshold value by minimizing the within-class variance. This method calculates the mean gray level values and variances of background and object pixels, and derives the threshold value. The threshold function can be implemented using Python and OpenCV, resulting in a binary image with white (building pixels) and black (background) pixels. Small sub-regions can be eliminated to improve the segmentation.

Accepted Answer

Contour approximation in building image processing aims to extract the rooftop/topmost point or pixel of the building roofline in the image. It generates an approximated contour containing all the quad corners (top, bottom, leftmost, rightmost). This method provides better accuracy compared to an RGB image and helps in localizing objects easily. By using contours, we can identify and analyze the building's structure effectively. Figure 7a and b illustrate the approximated contours of the building in the image, where blue, cyan, red, and green points out the top, bottom, leftmost, and rightmost border, respectively. Contour approximation is a crucial step in building image processing, as it enables the extraction of essential features and facilitates further analysis and interpretation of the building's characteristics.

Accepted Answer

Building footprint data is extracted using OpenStreetMap (OSM) and Nominatim. OSM provides geographical data on the Earth, while Nominatim is used for reverse geo-coding and geo-coding. By accessing the Nominatim API via HTTP request in Python, we can obtain latitude, longitude, place ID, OSM type, OSM ID, and bounding box. The OSM ID can be used to obtain the building footprint using OSMPython-Tools API. Nodes represent points on the map, and ways are ordered lists of nodes. Relations are ordered lists containing nodes, ways, or other relations. The extracted data helps in identifying the distance between the building and the camera for height estimation.

Accepted Answer

The pinhole camera projection framework can be used to calculate the height of a building by utilizing the camera coordinate system and the image plane coordinate system. The camera coordinate system consists of the origin O, and the x, y, and z axes, while the image plane coordinate system consists of the origin O and the x and y axes. By considering the distances from the camera to the building's roofline and floor, denoted as h_r and h_f respectively, and the focal length f of the camera, the height of the building can be calculated using the equation: h = (h_r * f) / d. This equation takes into account the Euclidean distance from the camera center to the building's nearer node, the focal length of the camera, and the projected length of the building's roofline on the image plane. By extracting the focal length from the image metadata and using the pinhole camera projection framework, the height of the building can be accurately determined.

Accepted Answer

In high-rise buildings, the camera pitch angle is set upward facing to capture the complete building's rooftop. However, due to the camera's movement, the error in height estimation may seem large. Despite this, the relative error is still admissible, as the error is relatively small compared to the building's height. For example, a 15 m error in estimation is only 3% of the relative error for buildings taller than 500 m. Therefore, while the absolute error may be high, the relative error remains acceptable for high-rise buildings.

Accepted Answer

The average relative error in height estimation for a 148m building is 2.0%. This was determined through an experiment where samples of random buildings were collected using Street View in North America. The samples were taken from different distant locations in the same direction by varying the camera-building distance. The relative error was plotted on a graph, showing a range between 0.3 to 3.15%. Despite the variation in distances, the average relative error remained low, indicating a reliable estimation method for high-rise buildings.

Accepted Answer

Varying directions can impact height estimation accuracy due to factors like urban density and vegetation. In the experiment, buildings were captured from different directions, maintaining similar distances. Figure 13b shows that the relative error in estimated height by varying directions ranges between 0.3 to 2.25% for a 148 m high building. This indicates that the variation in directions has a minimal effect on the accuracy of height estimation, making it a reliable method for estimating building heights in challenging scenarios.

Accepted Answer

For non-classification data types, accuracy is calculated by categorizing buildings based on their relative error. Buildings with less than 2% relative error are considered accurate, while the rest are errors. The accuracy is then determined by the percentage of accurate buildings. In this case, the accuracy was calculated for relative errors of 2%, 4%, 7%, and 10%, resulting in accuracies of 39%, 83%, 96%, and 96%, respectively. For an average case, the range of <4 to <10% relative error is considered, with accuracies ranging from 83% to 96%.

Accepted Answer

Normalized error is a statistical evaluation used to compare proficiency and performance measures of testing results. It determines the confirmation of the testing and can be evaluated using a specific equation. If the value results between -1 and +1, the result is acceptable; otherwise, the measurement needs to be reconsidered. Figure 14b represents the accuracy graph from the sample dataset, showing that the majority of building samples are estimated correctly, while a few exceed the range, indicating errors higher than the acceptable range. This suggests that some estimated results cannot be considered accurate.

Accepted Answer

In rural areas, vegetation and other objects can block the entire rooftop, making height estimation challenging. The DLIPHE algorithm struggles to detect and classify the edges of the facade in such cases. In dense urban areas, overlapping buildings make it difficult to detect boundaries in a 2D image. While capturing images from different angles can help, it may not always be feasible, especially in urban areas. For slanted high-rise buildings, estimating the approximate height becomes difficult. However, manipulating the pitch and distance of the camera in Google Street View API can slightly reduce the absolute error.

Accepted Answer

DLIPHE estimates building height by using deep learning and image processing techniques via the Street View API. It captures the building's image, applies deep convolutional neural network-based semantic segmentation to label the building's instance, and extracts the building's contour. OpenStreetMap data and the camera projection framework are then used to estimate the building's height. The methodology has been tested on low-rise and high-rise buildings, achieving decent accuracy and low relative error on high-rise buildings. Future work will focus on automating the process for high-rise buildings.

Data-Driven Diffraction Loss Estimation for Future Intelligent Transportation Systems in 6G Networks

Chat with Paper

AI Agents for this Paper

Most frequently asked questions

1. How can UAVs and eVTOLs improve communication in 6G networks?

2. What is the purpose of DLIPHE algorithm?

3. What factors affect diffraction loss estimation?

4. How do methods based on high-resolution optical imagery estimate building heights?

5. How is building height estimation using street-view data achieved?

6. How do UADs transmit data in urban environments?

7. What is the process of the DLIPHE model?

8. What parameters are required for Google Street View API request?

9. What is image segmentation?

10. What is image thresholding?

11. What is the purpose of contour approximation in building image processing?

12. How is building footprint data extracted?

13. How can the pinhole camera projection framework be used to calculate the height of a building?

14. How does camera pitch angle affect high-rise building height estimation?

15. What is the average relative error in height estimation for a 148m building?

16. How does varying directions affect height estimation accuracy?

17. How is accuracy calculated for non-classification data types?

18. What is normalized error in efficiency evaluation?

19. How do vegetation and overlapping buildings affect DLIPHE methodology?

20. How does DLIPHE estimate building height?

Citations

The Intelligent Traffic Flow Control System Based on 6G and Optimized Genetic Algorithm

Review on 6G Communication Network Technology for Sustainable Renewable Energy System: Implementations, Constraints, and Recommendations

DCTCADM: Data-Centric Cognitive Agent Decision Making Based an Effective Handoff for Heterogeneous Wireless Communication

References

Deep Residual Learning for Image Recognition

Xception: Deep Learning with Depthwise Separable Convolutions

Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation

The Cityscapes Dataset for Semantic Urban Scene Understanding

OpenStreetMap: User-Generated Street Maps

Related Papers (5)

Path loss models for outdoor environment—with a focus on rain attenuation impact on short-range millimeter-wave links

Radiowave Propagation Over an Irregular Surface – The GO and UTD Approach

Material Characterisation for Short Range Indoor Environment in the Millimetre Wave Bands

Measurements of path loss in MM-wave for indoor environments

Break Point Distance Path Loss Model for Indoor Wireless Communications with Equivalent Floor