Evaluating Urban Greenery Through the Front-Facing Street View Imagery: Insights from a Nanjing Case Study

Jin Zhu, Yingjing Huang, Ziyue Cao, Yue Zhang, Yuan Ding, Jinglong DuISPRS Int. J. Geo Inf 2025 · DOI

TL;DR

This paper introduces two new metrics, FFGVI and SCGVI, to better assess urban greenery from a pedestrian's perspective using street view imagery. These metrics offer a more nuanced understanding of urban green spaces than traditional methods, particularly in dense urban environments, and can inform urban planning and tourism.

What Problem Does This Paper Address?

Urban green spaces are crucial for city livability, offering numerous environmental and health benefits. Quantifying urban greenery is essential for effective urban planning and management. Traditional methods often rely on satellite imagery or broad-scale metrics. More recently, street view imagery has been used to calculate the Green View Index (GVI), which measures the proportion of green pixels in an image. However, the standard GVI, often calculated from images taken in four cardinal directions, doesn't fully capture the human-scale experience of greenery. Pedestrians primarily orient their gaze forward, and the GVI can be influenced by factors like the viewing angle and obstructions, potentially overestimating or underestimating perceived greenery. This paper addresses the need for metrics that better reflect a pedestrian's visual experience of urban greenery, especially in complex street environments.

Methods

The study introduces two novel metrics:

  • Front-Facing Green View Index (FFGVI): This metric focuses specifically on the greenery visible in front-facing street view images, mimicking a pedestrian's direct line of sight. It's calculated as the proportion of green pixels in the entire front-facing image.
  • Street Canyon Green View Index (SCGVI): This metric is designed to identify "boulevards" that create a sense of enclosure and aesthetic appeal. It's calculated using only the upper half of the front-facing street view image, emphasizing how tree canopies obscure the sky and contribute to a "room-like" streetscape.

The methodology involves:

  1. Azimuth Calculation: Determining the precise heading (azimuth) for each sampling point using OpenStreetMap (OSM) road network data. This is a more efficient approach than relying on image metadata.
  2. Street View Image Acquisition: Retrieving front-facing street view images from the Baidu Maps platform using the calculated azimuths.
  3. Semantic Segmentation: Employing the MP-Former model, a deep learning-based semantic segmentation framework, to accurately classify pixels into categories like vegetation, sky, and buildings. This is a crucial step for identifying green pixels.
  4. Metric Calculation: Computing FFGVI and SCGVI based on the segmented images.
  5. Analysis: Evaluating the distributions and correlations of FFGVI, SCGVI, and the traditional GVI.

Datasets

  • OpenStreetMap (OSM) road network data: Used for generating sampling points and calculating azimuths.
  • Baidu Street View Images (BSVI): The primary source of imagery for the study. The authors note that Google Street View (GSV) could also be used.
  • Cityscapes dataset: Used for training and evaluating the MP-Former semantic segmentation model.

Key Insights and Ideas

  • Pedestrian Perspective Matters: The study strongly advocates for a pedestrian-centric approach to assessing urban greenery, highlighting that a forward-facing view is more representative of daily experience than multi-directional views.
  • GVI Overestimation: Traditional GVI, especially when derived from four directions, can overestimate perceived greenery because side-facing views often capture more trees closer to the camera, leading to a "proximity-to-camera effect."
  • FFGVI as a Human-Centric Alternative: FFGVI offers a more accurate reflection of what pedestrians actually see and experience, mitigating biases from sampling point locations and obstructions like large vehicles.
  • SCGVI for Boulevard Identification: SCGVI effectively identifies boulevards that provide a sense of enclosure and aesthetic appeal, contributing to a positive urban experience and potentially mitigating solar glare.
  • Efficiency Gains: Using only front-facing images reduces data volume by a factor of four compared to traditional GVI methods, streamlining storage and processing.

Key Results

  • Strong Correlation: FFGVI showed a strong positive correlation with GVI (R = 0.88).
  • Moderate Correlation: SCGVI had a weaker, though still significant, correlation with GVI (R = 0.78).
  • High Correlation between FFGVI and SCGVI: These two metrics, both derived from front-facing images, exhibited a very strong correlation (R = 0.91).
  • GVI vs. FFGVI/SCGVI: GVI values generally tended to be higher than FFGVI and SCGVI, supporting the idea of GVI overestimation.
  • Boulevard Identification: The SCGVI metric successfully identified ten boulevards in Nanjing with SCGVI values above 90%, characterized by dense tree canopies and a sense of enclosure.
  • Robustness: The study demonstrated that FFGVI is less sensitive to sampling point location variations and large vehicle occlusions compared to traditional GVI.

Limitations and Future Work

  • Semantic Segmentation Accuracy: While MP-Former achieved good accuracy, minor segmentation errors were observed, potentially leading to slight overestimation of SCGVI in some cases where sky gaps within foliage are misclassified as greenery.
  • Street View Image vs. Sidewalk Experience: The study acknowledges that street view images, captured from vehicles, may differ slightly from the experience of pedestrians walking on sidewalks, especially on wider roads.
  • Contextual Factors: The influence of factors like average tree spacing and road width on GVI variability warrants further investigation.
  • Future Work: The authors suggest that FFGVI and SCGVI can be integrated with network flow algorithms for optimizing route planning, particularly for scenic tourist routes. They also propose further research into the factors influencing GVI variability.

Summary updated

BibTeX
@article{zhu2025evaluating,
  author = {Jin Zhu and Yingjing Huang and Ziyue Cao and Yue Zhang and Yuan Ding and Jinglong Du},
  title = {{Evaluating Urban Greenery Through the Front-Facing Street View Imagery: Insights from a Nanjing Case Study}},
  journal = {ISPRS Int. J. Geo Inf.},
  year = {2025},
  doi = {10.3390/ijgi14080287},
}
Figure