[Paper Review] ELEV-VISION: Automated Lowest Floor Elevation Estimation from Segmenting Street View Images

Yu-Hsuan Ho, Cheng-Chun Lee+3 more|arXiv (Cornell University)|Jun 5, 2023

Flood Risk Assessment and ManagementEnvironmental Science3 citations

TL;DR

This paper proposes ELEV-VISION, an automated method for estimating lowest floor elevation (LFE) and height difference between street and lowest floor (HDSL) using image segmentation on Google Street View panoramas. By leveraging depthmaps and pitch angle calculations from equirectangular projections, it achieves a mean absolute error of 0.190 m (1.18%) in LFE estimation, offering a cost-effective, high-resolution alternative to manual surveys or drone-based methods.

ABSTRACT

We propose an automated lowest floor elevation (LFE) estimation algorithm based on computer vision techniques to leverage the latent information in street view images. Flood depth-damage models use a combination of LFE and flood depth for determining flood risk and extent of damage to properties. We used image segmentation for detecting door bottoms and roadside edges from Google Street View images. The characteristic of equirectangular projection with constant spacing representation of horizontal and vertical angles allows extraction of the pitch angle from the camera to the door bottom. The depth from the camera to the door bottom was obtained from the depthmap paired with the Google Street View image. LFEs were calculated from the pitch angle and the depth. The testbed for application of the proposed method is Meyerland (Harris County, Texas). The results show that the proposed method achieved mean absolute error of 0.190 m (1.18 %) in estimating LFE. The height difference between the street and the lowest floor (HDSL) was estimated to provide information for flood damage estimation. The proposed automatic LFE estimation algorithm using Street View images and image segmentation provides a rapid and cost-effective method for LFE estimation compared with the surveys using total station theodolite and unmanned aerial systems. By obtaining more accurate and up-to-date LFE data using the proposed method, city planners, emergency planners and insurance companies could make a more precise estimation of flood damage.

Motivation & Objective

• Address the lack of accurate, up-to-date LFE data for flood damage modeling and urban resilience planning.
• Overcome limitations of traditional LFE measurement methods such as total station theodolite surveys, which are costly and time-consuming.
• Improve upon prior image-based LFE estimation methods that rely on object detection with bounding boxes, which suffer from distortion and information loss during reprojection.
• Enable estimation of HDSL—the height difference between street level and the lowest floor—as a critical indicator for flood vulnerability assessment.
• Develop a scalable, automated, and cost-effective method for large-scale LFE and HDSL estimation using publicly available street view imagery and computer vision.

Proposed method

• Collect street view panoramas from Google Street View using a predefined street map as the data source.
• Apply semantic image segmentation to detect door bottoms and roadside edges directly on equirectangular street view panoramas, preserving spatial integrity without reprojection.
• Extract the pitch angle from the camera to the door bottom using the equirectangular projection’s constant angular spacing in horizontal and vertical dimensions.
• Obtain depth information from paired depthmaps (256×512 resolution) corresponding to each street view image.
• Calculate LFE using the formula: LFE = camera height + (depth × sin(pitch angle)) + (height offset from door bottom to lowest floor), where the height offset is derived from segmentation masks.
• Estimate HDSL as the vertical difference between the segmented roadside and the door bottom, enabling flood damage modeling at the building scale.

Experimental results

Research questions

• RQ1 Can image segmentation on equirectangular street view panoramas enable accurate and robust LFE estimation without image reprojection or bounding box limitations?
• RQ2 How does the proposed method compare in accuracy to existing LFE estimation techniques using object detection or manual surveys?
• RQ3 To what extent does viewpoint variation or image distortion affect the performance of the segmentation-based LFE estimation?
• RQ4 Can the method reliably estimate HDSL—the height difference between street level and the lowest floor—for improved flood damage modeling?
• RQ5 What are the key limitations of the method, particularly regarding depthmap resolution and visibility of the front door?

Key findings

• The proposed ELEV-VISION method achieved a mean absolute error (MAE) of 0.190 m (1.18%) in LFE estimation on the Meyerland testbed in Harris County, Texas.
• The method demonstrated robustness to partial occlusion of the door bottom, with no statistically significant impact on estimation accuracy (p-value > 0.1).
• Image segmentation outperformed bounding box-based object detection by preserving shape fidelity under viewpoint variation and projection distortion, especially in curved or non-rectilinear scenes.
• The method successfully estimated HDSL without relying on assumed reference heights, reducing error propagation compared to prior approaches.
• The depthmap resolution (256×512) limits the method’s accuracy for distant objects, as depth differences below 0.7 degrees may not be captured.
• The method is currently constrained to buildings with visible front doors; future work could extend it to other features like windows or stairs, though with increased uncertainty.