CityPulse: Fine-Grained Assessment of Urban Change with Street View Time Series

TL;DR

This paper tackles the problem of fine-grained urban-change assessment by introducing the largest street-view time-series dataset to date and an end-to-end change-detection pipeline. It leverages a Siamese network with a DINOv2 backbone to learn from time-series street-view data and detect urban change points across cities, demonstrating city-scale applicability with a Seattle case study. The key contributions include a large multi-city street-view time-series dataset with image-level change labels, an end-to-end pipeline that outperforms pairwise baselines, and evidence that detected change points correlate with socio-demographic shifts beyond what construction permits proxies capture. The work provides a practical, high-resolution visual proxy for urban evolution that can support urban planning and policy evaluation at finer spatial and temporal granularity.

Abstract

Urban transformations have profound societal impact on both individuals and communities at large. Accurately assessing these shifts is essential for understanding their underlying causes and ensuring sustainable urban planning. Traditional measurements often encounter constraints in spatial and temporal granularity, failing to capture real-time physical changes. While street view imagery, capturing the heartbeat of urban spaces from a pedestrian point of view, can add as a high-definition, up-to-date, and on-the-ground visual proxy of urban change. We curate the largest street view time series dataset to date, and propose an end-to-end change detection model to effectively capture physical alterations in the built environment at scale. We demonstrate the effectiveness of our proposed method by benchmark comparisons with previous literature and implementing it at the city-wide level. Our approach has the potential to supplement existing dataset and serve as a fine-grained and accurate assessment of urban change.

Figures (7)

• Figure 1: Detection of urban change points using street view time series. Red bounding boxes highlight transformations in the built environment at each location. By aggregating these detected change points within a neighborhood, we can evaluate the temporal dynamics of urban development.
• Figure 2: Geo-spatial distribution of our street view time series dataset across 5 different cities in the US. Locations are selected based on open-access building footprint data, and historical Google Street View imagery from these coordinates is comprehensively downloaded and labeled with urban change points.
• Figure 3: Partitioning of street view time series data. All possible pairwise combinations of street view samples are generated from each time series. Each pair's label is assigned based on its corresponding position with the urban change points.
• Figure 4: Overview of the change detection model architecture. Pairs of input images are processed using Siamese-based networks with DINOv2 as the backbone. The CLS tokens serve as the image representation, with a subsequent linear layer projecting them to a prediction score.
• Figure 5: Sampled prediction results. Our proposed change detection model effectively identifies structural changes in buildings, while filtering our random variations such like lighting, shadows, vegetation, and vehicles.

Theorems & Definitions (2)

• Definition 1
• Definition 2