| 摘要: |
| 以街景影像为代表的人视角景
观环境量化，是当前研究人与城市环境
互动关系的热门方法。然而现有主流视
觉量化受二维影像制约存在透视畸变与
信息缺失等问题。基于此，本文首先从
理论层面分析当前主流视觉量化方法存
在的问题，提出基于“角”的人视角环
境基础空间计量；通过实验验证其较传
统方法反映真实视觉感知的差异性；研
发量化工具Env View. P，以绿视率为
例，展示基于该计量的量化方法在环境
视觉量化过程中的具体应用。研究表
明，基于“角”的计量方法能够显著提
升绿视率等指标的效度与信度，高效的
工具实现了场景模拟与个体语义识别的
扩展能力。研究构建了更符合真实视觉
的人视角量化体系，为风景园林及相关
领域提供了可靠的空间计量基础与研究
语料。 |
| 关键词: 环境视觉 绿视率 计算机视
觉 街景影像 数字景观技术 |
| DOI：10.13791/j.cnki.hsfwest.20241023001 |
| 分类号: |
| 基金项目:国家自然科学基金青年基金项目（52508093、52308084）；河南省科技攻关项目（252102320288、242102320309） |
|
| Spatial metrics and quantification of environment at eye level: A green view index casestudy |
|
ZHANG Yi,FU Mengze,LUAN Chunfeng
|
| Abstract: |
| Quantifying the built and natural environment from the human-eye perspective has become a
widely used approach in landscape architecture, urban design, environmental psychology, and related fields.
With the increasing availability of street-view imagery and advances in computer vision, large-scale humanperspective
environmental data can now be routinely extracted and analyzed. However, most existing visual
quantification methods rely on two-dimensional image space and represent environmental components by
pixel proportions. These methods overlook the fact that human visual perception is based on spherical
retinal imaging rather than planar projection, thus creating inconsistencies between computational
measurements and actual perceptual experience. Furthermore, pixel-based indicators are sensitive to
projection distortions, highly dependent on camera parameters, and incapable of representing depth
structures or distinguishing individual landscape elements. As these methods form the basis of numerous
data-driven empirical studies, their limitations introduce potential biases and restrict the scientific rigor and
application scope of human-perspective environmental research. To address these issues, this study proposes
a basic spatial measurement system for human-perspective environmental quantification grounded in
angular geometry. Building on theories of retinal projection and spherical visual space, we develop a
measurement framework that replaces pixel proportions with angular units that more accurately describe the
geometric properties of the visual field. Specifically, the system consists of four components: 1) onedimensional
angular measurement using radians to quantify visual position, height, width, and boundary
features; 2) two-dimensional measurement using solid angles to represent the perceived “area” of visual
objects on the spherical retina; 3) depth-related representation using relative distance relationships to
capture occlusion, spatial layering, and volumetric order; and 4) individual semantic labeling using unique
identifiers to distinguish different objects and enable object-level visual impact analysis. These components
collectively define a perceptually grounded, geometrically meaningful basis for human-perspective
environmental quantification. To evaluate the alignment between different quantification methods and real
human perception, the study designed a psychophysical experiment using paired spherical objects.
Participants viewed a reference sphere placed at eye level and judged the distance at which a second
identical sphere appeared to have half of its visual diameter. Across a range of elevation angles, it recorded
the subjective equivalence distances and compared them against measurements obtained from multiple
computational methods, including cube mapping, equirectangular projection, fisheye projection, and the
proposed angular metrics. Experimental data from 20 participants show that angular metrics—both radianbased
and solid-angle-based—produce one-dimensional and two-dimensional size ratios closest to the
perceptual benchmark. They outperform all planar projection-based methods in both validity and stability.
Analyses further confirm that fisheye projection behaves similarly to angular measurement in many cases,
which is consistent with its geometric properties, but still exhibits elevation-dependent variation that angular
measurement does not. Based on these theoretical and experimental results, we take the visible green index
(GVI) as a representative application to demonstrate the implications of adopting angular metrics inpractical landscape analysis. Using a virtual urban plaza model, we computed GVI using the proposed solid-angle method and compared it with GVI derived
from conventional equirectangular panoramic imagery. Results show systematic differences: pixel-based GVI consistently underestimates visible greenery, with
errors reaching up to 10% in some scenarios. The largest deviations occur in the quantification of trees, whose canopies typically occupy higher elevation
angles and thus are more susceptible to projection distortions. Shrubs and grass exhibit relatively smaller deviations. Spatial interpolation of GVI values further
reveals that the magnitude and direction of errors vary with viewing distance and vertical distribution of vegetation, reflecting the nonuniform distortions
inherent in planar projections. To support efficient and scalable implementation of the proposed method, we developed Env View.P, an open-source tool that
integrates the angular measurement framework with a ray-tracing-based spherical sampling workflow. Env View. P converts polygonal mesh models—
commonly used in landscape design and simulation—into depth and semantic matrices through dense ray sampling, performs uniform resampling using
Fibonacci sphere grids, computes solid angles for each sampling direction, and outputs multi-dimensional environmental indices. The tool is capable of
generating high-resolution multi-layer data in seconds, allowing for large-scale analysis, scenario simulation, limited-view analysis, and object-level visual
contribution mapping. Demonstration scenarios show its ability to quantify multiple green-view components, evaluate directional visibility, and assess depthlayered
vegetation structures, thereby supporting more nuanced design decisions, visibility research, and environmental assessment tasks. Overall, this study
establishes a perceptually and geometrically principled framework for human-perspective environmental quantification, validates its accuracy and stability
through controlled experiments, and provides an efficient tool for practical application. The findings confirm that angular metrics better represent real visual
perception compared with traditional pixel-based approaches, and that adopting such metrics enhances the accuracy and interpretability of GVI and related
indicators. The methodology and tool extend the analytical capability of human-perspective landscape studies, supplying reliable geometric foundations and
richer data for subsequent research on environmental aesthetics, health-supportive landscapes, embodied perception, environmental equity, and other topics
requiring rigorous visual quantification. |
| Key words: environment vision visible green index computer vision street view imagery digital landscape technology |
