Research on Urban Street Spatial Quality Based on Street View Image Segmentation

Abstract

Assessing the quality of urban street space can provide suggestions for urban planning and construction management. Big data collection and machine learning provide more efficient evaluation methods than traditional survey methods. This study intended to quantify the urban street spatial quality based on street view image segmentation. A case study was conducted in the Second Ring Road of Changsha City, China. Firstly, the road network information was obtained through OpenStreetMap, and the longitude and latitude of the observation points were obtained using ArcGIS 10.2 software. Then, corresponding street view images of the observation points were obtained from Baidu Maps, and a semantic segmentation software was used to obtain the pixel occupancy ratio of 150 land cover categories in each image. This study selected six evaluation indicators to assess the street space quality, including the sky visibility index, green visual index, interface enclosure index, public–facility convenience index, traffic recognition, and motorization degree. Through statistical analysis of objects related to each evaluation indicator, scores of each evaluation indicator for observation points were obtained. The scores of each indicator are mapped onto the map in ArcGIS for data visualization and analysis. The final value of street space quality was obtained by weighing each indicator score according to the selected weight, achieving qualitative research on street space quality. The results showed that the street space quality in the downtown area of Changsha is relatively high. Still, the level of green visual index, interface enclosure, public–facility convenience index, and motorization degree is relatively low. In the commercial area east of the river, improvements are needed in pedestrian perception. In other areas, enhancements are required in community public facilities and traffic signage.

1. Introduction

As of 2020, about 56% of the global population lived in urban areas [1]. In urban construction, street space quality and vitality are the cornerstone of improving the living environment of streets. The expectations of people for urban roads are no longer solely based on their basic transportation function but also on the desire to enhance the awareness of the harmonious relationship between the environment, psychology, and behavior of road users, thereby forming a humanized urban road system [2]. People-centered urban planning is attracting more attention, requiring planners to consider urban issues from a more refined and humane perspective. The urban street space is where people gather for communication and activities. It is one of the essential indicators that could reflect the quality of the city [3]. Within this context, the focus of this study is to evaluate the quality of urban street spaces in a manner that is not only rapid and convenient but also comprehensive and integrative.

In the study of street space quality, many scholars prioritize the human scale as the basis [4]. Traditional survey methods commonly use on-site observation and questionnaires to gather honest feedback and conclusions, summarizing the required research features and findings [5]. Zhang et al. [6] evaluated the spatial carrying capacity, street attractiveness, travel safety, environmental comfort, and social interaction of Shuanggang Old Street in Hefei City using a questionnaire survey method. They found that travel safety and social interaction significantly affect the quality of commercial streets. Itma and Monna [7] developed a questionnaire for the collective space development of residential areas in Palestine, measuring the satisfaction of people from the perspective of social sustainability values. To explore the influence of street space visual quality on human physiology and comfort perception, Gorgul et al. [8] compared the heart rate and comfort survey questionnaire of 15 participants when looking at a crossroads and a street. Traditional questionnaire survey methods rely on the extensive collection of participant feedback for analysis. This process is not only time-consuming but also susceptible to the subjectivity of the respondents, which may lead to delays and a lack of objectivity in research conclusions.

Integrating information technologies such as the Internet of Things, cloud computing, big data analysis, and spatial geography information facilitates the planning and management of urban development [9,10]. In urban development, it is crucial to focus on reducing energy consumption and improving environmental quality in order to achieve sustainable development. Utilizing advanced urban energy models and other tools enables a rapid understanding of urban energy demand, providing robust theoretical support for enhancing building energy efficiency, decreasing carbon emissions, and creating a more comfortable urban environment [11,12,13]. It leads to the creation of smart cities. This efficient urban operation and service mechanism provides people with more convenient living services [14]. Jiang et al. [15] analyzed the relationship between street vitality and building environmental characteristics in 14 Chinese cities based on urban expansion remote-sensing monitoring databases and population density data to expand the scope of information acquisition and improve research efficiency. Sigala et al. [16] developed a smartphone application that uses sensors in built-in devices to detect the surface quality of urban streets, achieving large-scale and continuous data collection. In addition, Huang and Wang [17] detected urban structures in space and time dimensions using big spatial data (BSD), nighttime satellite light images, and social media check-in data. These research findings not only enrich the theoretical framework for assessing urban spatial quality but also offer new perspectives and methodologies for urban planning and quality management in practice.

With the development of big data analysis and machine learning, remote virtual evaluation methods represented by street view images (SVIs) are receiving increasing attention and application [18,19,20,21]. Machine learning can quickly identify elements in the image and combine them with relevant evaluation indicators to achieve street space quality assessment. The convolutional neural network (CNN) model has an encoder–decoder architecture that can extract image feature data and convert them into semantic segmentation results through the decoder, effectively solving the problem of multi-class pixel classification in semantic segmentation tasks, thus improving segmentation accuracy and speed [22,23]. Dai et al. [24] used the fully convolutional network (FCN-8s) to segment Wuhan street scene images. Weng et al. [25] proposed a deep multi-branch aggregation network (DMA-Net) based on the encoder–decoder structure for real-time semantic segmentation in street scenes, achieving a good balance between accuracy and speed. Wang et al. [2] used an improved semantic image segmentation method based on the encoder–decoder structure, SegNet, to perform semantic classification on 126,724 SVIs of Binjiang District, Hangzhou, Zhejiang Province. To explore the advantages and disadvantages of different deep-learning models in urban perception prediction tasks, Wang et al. [26] compared models based on convolutional neural networks (CNNs) and models based on fully convolutional neural networks and random forests (FCNs + RFs) and found that the model based on FCNs + RFs is suitable for scenes with substantial spatial heterogeneity (such as the central urban areas of the megacities of China). Changsha City is a significant economic hub within Hunan Province and a central city in the middle reaches of the Yangtze River Economic Belt [27]. The area within the second ring road of Changsha encompasses numerous tourist attractions, green parks, and core commercial districts [28]. Based on the aforementioned analysis, this article chose a semantic segmentation model based on FCN to identify elements in SVIs. By analyzing the quality of street spaces within the second ring road of Changsha, this study not only synthesizes a reflection of the distinctive characteristics of streets in various urban areas but also provides important references for future urban road planning and construction.

Based on the definition of street spatial quality measurement methods, the system mainly focuses on the physical environment and pedestrian perception [29]. Physical features are exciting and important parts or features of street elements that people can see [30]. Choosing different streetscape features as street quality evaluation indicators will produce different results [30,31]. In the physical environment, the evaluation system involves street functions, diversity, and serviceability. Millstein et al. [32] studied the impact of streetscape characteristics on physical activities and found that frontal streetscape projects that affect walking are related to pedestrian signs and street facility equipment. Yu et al. [33] found that traffic signs are significant in guiding driving behavior and ensuring the convenience and safety of traffic activities. They can also provide the decisive basis for various intelligent transportation systems.

In measuring pedestrian perception, a comprehensive multi-sensory measurement based on sensory stimulation and environmental cognition is used in academic discussions and applications. The dimensions of the measurement perspective are divided into five categories: psychological, visual, auditory, tactile, and olfactory [29]. Tao et al. [34] found that high green visibility, high sky visibility, and natural revetment significantly impact public sentiment. Through regression analysis, Akbarishahabi et al. [35] tested the relationship between the enclosure of street ratio and street skyline fractal dimension. They proposed an objective evaluation method to evaluate the enclosure ratio of the street. Therefore, to comprehensively measure street quality, applying the multi-indicator evaluation was necessary. Zhang et al. [36] selected density, diversity, design, destination accessibility, and distance to transit as five key variables that affect street quality. However, they did not take into account the impact of pedestrian perception on street quality. Duan et al. [37] selected six-dimensional indicators based on the perspectives of residents: convenience, functionality, safety, comfort, sociality, and visualization, as an evaluation system that reflects the characteristics of living streets. This approach measures street quality from the subjective viewpoint of residents, lacking objectivity. Wu et al. [38] believed that the critical design elements that affect the experience of the visual quality of street landscapes include diversity, building frontage, enclosure, greenery, street width, pedestrian space, motorization, and sky views, covering various factors that affect the visual quality of street landscapes but failing to consider street functionality, such as the importance of traffic signs.

An analysis of prior research indicates that a comprehensive evaluation of street space quality should originate from both the physical environment and pedestrian perception, taking into account the functionality, visual comfort, serviceability, and convenience of the streets. Street facility equipment, traffic signs, and the degree of motorization are important physical characteristics for assessing the quality of street spaces. Green visibility, sky visibility, and the enclosure ratio of the street are significant factors affecting pedestrian perception.

In summary, research on small-scale street public spaces was limited by various factors such as data conditions and technical levels. However, the development of big data technology has become a new technological means for obtaining data and conducting research and analysis in the planning field, as it can more accurately reflect the behavioral characteristics of people. This paper selects the partial urban roads within the second ring in Changsha City, China, as the research object. Based on SVIs, it utilizes semantic segmentation and ArcGIS, and selects evaluation indicators to construct a street space quality measurement system. The system rapidly acquires pixel proportion data of 150 land cover categories from a large number of SVIs and visualizes the scores of street space under different indicators. We explored the quality of the city street space through six evaluation indicators, including the sky visibility index (SVI), green visual index (GVI), interface enclosure index (IEI), public–facility convenience index (PCI), traffic recognition (TR), and motorization degree (MD). Improvement strategies for urban architectural planning, greening, public facilities, and road construction will be based on the evaluation results. Therefore, this study provides a reference for dynamic monitoring and management of street space quality.

The remainder of this paper is organized as follows: Section 2 introduces the research area and the methods for obtaining and identifying SVIs, focusing on the calculation methods of six evaluation indicators and the method for assessing the overall street spatial quality. Section 3 presents the evaluation results of the research area based on physical environment and pedestrian perception, as well as the six factors, along with the comprehensive evaluation results of street spatial quality. Section 4 discusses the shortcomings of the study. Section 5 provides the conclusions of this research.

2. Methodology

The main goal of this study is to evaluate the spatial quality of urban streets through image recognition technology. Therefore, our research is divided into two main phases: image recognition and street quality assessment. Figure 1 below displays the process of urban street space quality research based on image recognition. In the data processing and image recognition stage, we first determined the research scope and obtained the road network information within this scope. Next, we determined the observation points and obtained the SVIs of the observation points. Then, we used a semantic segmentation recognition software called GPU-CUDA-enabled Semantic Segmentation App. v1.0 to identify the features and the pixel proportions of various features in each SVI. This process generates a CSV file containing all the data we need. In the street quality evaluation stage, we selected six key indicators, SVI, GVI, IEI, PCI, TR, and MD, to form the evaluation system for urban street space quality. We performed a statistical analysis of the features related to each indicator and performed weighted statistics on each indicator to achieve a comprehensive and detailed assessment of street quality. Finally, we used ArcGIS 10.2 software to perform a hierarchical visualization of each indicator data. The final evaluation results were obtained through the hierarchical visualization and comprehensive evaluation and analysis of the quality of urban street space.

2.1. Data Gathering and Processing

2.1.1. Gathering the Data of the SVIs

The scope of this study is the area within the Second Ring Road of Changsha City. As shown in Figure 2, the area in the yellow dashed line is the study area. The area is westbound to the West Second Ring, southbound to the South Second Ring, northbound to Yuelu Avenue and Sanyi Avenue, and eastbound to the East Second Ring. The area is 88.40 km², covering different administrative and planning areas of the new and old districts. There are many types of roads within the study area, including main roads, secondary roads, branch roads, and ring roads, which can comprehensively reflect the quality of street space in the Changsha urban area.

Vector road network information can be obtained through the OpenStreetMap website. Short, dense, and atypical roads were removed. Simplify the road network and merge the two road lines into one line to avoid repeated acquisition of observation points. Based on the effective visual range of urban streets and the complexity of the roads, an observation interval of 200 m was chosen. As illustrated in Figure 3, ArcGIS generated 16,687 observation points in the area. According to the latitude and longitude coordinates of the observation points, the Baidu Street View Image Application Programming Interface was used to obtain SVIs [28]. To better simulate the pedestrian perspective and make the visibility range cover the horizontal and vertical directions of each observation point, set the parameter of the perspective range to 360°. A total of 10,666 images were obtained. Due to file corruption and image distortion during the transmission process, only 8047 valid pictures were obtained. The image pixels were 1024 × 512. They were taken in August 2019. As shown in Figure 4, the image quality is high and can clearly and intuitively represent the street space from the perspective of pedestrians. The locations of obtained SVIs are distributed in the streets within the study area. By semantically segmenting the SVIs within each administrative region and conducting visual analysis using ArcGIS, the performance of different administrative regions in the six evaluation indicators can be intuitively observed.

2.1.2. Extraction of Street View Elements

We identified various elements in the SVIs through semantic segmentation and obtained the pixel proportion data of each element in the image. Then, the deep learning complete convolutional network (FCN) based on the ADE_20K dataset trained by Yao et al. [39] was selected for semantic segmentation of the SVIs. This model has a pixel comparison accuracy of 0.814 on the training dataset and 0.668 on the test dataset. Figure 5 illustrates the process of identifying SVIs using the FCN model. Semantic segmentation aims to segment the input image based on semantic information and predict the semantic category of each pixel from a given set of labels [30]. The label set of the image semantic segmentation software includes 150 types of elements, such as sky, plants, buildings, cars, and traffic lights. As shown in Figure 5, it can be displayed in ArcGIS after the image is processed by semantic segmentation. Different colors represent each element; the color boundaries are clearly distinguished and highly consistent with the original image. At the same time, we obtained the pixel proportion data of each element in the image. These data are fundamental for us to evaluate the quality of street space. For example, SVI and GVI can be assessed by analyzing the pixel ratio of sky and plant elements. The software has a user-friendly interface and efficient running speed.

2.2. Construct Evaluation Indicators

Assessing the quality of street space is a complex process that requires consideration of multiple factors. This paper selected the two dimensions of physical environment and pedestrian perception as the measurement system for street spatial quality. Among them, the physical environment includes the PCI, TR, and MD, and pedestrian perception includes the SVI, GVI, and IEI. These factors will have varying degrees of impact on crowd activities and interactions between people.

The SVI is an essential indicator for the evaluation of street space quality. A good SVI can make people feel comfortable on the street and enjoy natural light while also helping to improve the aesthetics of the street. The GVI is essential for evaluating street space quality. The GVI reflects the degree of coverage of green plants on the street. A high GVI can make the street environment more pleasant and help improve the quality of life. The IEI, PCI, TR, and MD are essential indicators for evaluating street space quality. The degree of IEI reflects the degree of enclosure of the street space by buildings on both sides. A high IEI can make the street space more compact and help improve the activity efficiency of residents. The degree of PCI reflects the quantity and quality of public facilities on the street. A high PCI can make the activities of people on the street more convenient. TR reflects the awareness of people of street traffic conditions. High TR can make people drive safer on the streets. The MD reflects the number and speed of motor vehicles on the streets. A high MD can make people drive faster on the streets.

Based on the above six factors, we can comprehensively evaluate the street space quality of specific areas within the Second Ring Road of Changsha City. We refer to the indicators shown in

Table 1. Criteria of each evaluation element.

Evaluation Rank	Low	Relatively Low	Relatively High	High
SVI	≤10%	10~15%	15~20%	≥20%
GVI	≤5%	5~10%	10~15%	≥15%
IEI	≤15%	15~30%	30~45%	≥45%
PCI	≤1%	1~2%	2~3%	≥3%
TR	≤0.5%	0.5~1%	1~1.5%	≥1.5%
MD	≤15%	15~30%	30~45%	≥45%
Street spatial quality	≤9%	9~12%	12~15%	≥15%

for evaluation and weigh each element according to the weight shown in

Table 2. Weights of the six elements.

Establishing Dimensions	Indicators	Weights	Related Attributes
Pedestrian Perception	SVI	0.2090	Positive correlation
	GVI	0.1836	Positive correlation
	IEI	0.1995	Negative correlation
Physical Environment	PCI	0.0488	Positive correlation
	TR	0.1566	Positive correlation
	MD	0.2025	Negative correlation

to obtain the final value of street space quality. In this way, we can more comprehensively and accurately evaluate the street space quality of specific areas within the Second Ring Road of Changsha City and provide a reference for future urban planning and construction.

2.2.1. Sky Visibility Index (SVI)

The SVI measures the proportion of sky that people can see in urban open spaces. This indicator depends to a large extent on the development layout of urban construction. In urban spaces surrounded by buildings, the SVI directly affects the comfort of urban public activities. Therefore, it is considered an important indicator for evaluating the quality of street space. In this research, some specific land species to study SVI were chosen. Based on the types of objects identified by semantic segmentation, this study selected the sky and airplanes as the subjects for analyzing SVI. The sum of the pixel proportions of these two categories represents the SVI of the image, calculated using Equation (1):

$$SVI={\displaystyle \frac{P_{\mathrm{sky}}+P_{airplane}}{P_{all}}}$$

(1)

2.2.2. Green Visual Index (GVI)

The GVI refers to the proportion of greenery in street spaces determined by the plants within the space. The GVI reflects the degree of greenery in the neighborhood, which contributes to the comfort level of its inhabitants. Trees, grass, and other plants were selected as the research objects for the street GVI, which is calculated using Equation (2):

$$GVI={\displaystyle \frac{P_{tree}+P_{grass}+P_{plant}}{P_{all}}}$$

(2)

2.2.3. Interface Enclosure Index (IEI)

The IEI refers to the degree of enclosure of the street space by the facades on both sides of the road, which is determined by the buildings and plantings on both sides of the street. The IEI of street spaces reflects the scale of openness of streets. Walls, buildings, houses, trees, windows, grass, plants, fences, skyscrapers, and screen doors were selected as the subjects that contribute to the IEI. The sum of the pixel proportion in the image for these ten types of objects represents the IEI of the image, calculated using Equation (3):

$$IEI={\displaystyle \frac{P_{wall}+P_{building}+P_{house}+P_{tree}+P_{window}+P_{grass}+P_{plant}+P_{fance}+P_{skyscraper}+P_{screen door}}{P_{all}}}$$

(3)

2.2.4. Public–Facility Convenience Index (PCI)

The PCI refers to the proportion of various facilities in the street space, including seating, streetlights, trash cans, traffic signs, and public restrooms. The facility of a block is an essential condition for regional construction and development, reflecting the infrastructure level of the block. Chairs, toilets, benches, stools, trash cans, tables, armchairs, lights, streetlights, and bulletin boards were selected as the subjects of PCI. The sum of the pixel proportion for these ten types of objects represents the PCI of the image, calculated using Equation (4):

$$P_{CI}={\displaystyle \frac{P_{chair}+P_{toilet}+P_{benche}+P_{stool}+P_{trash can}+P_{table}+P_{armchair}+P_{light}+P_{streetlight}+P_{bulletin board}}{P_{all}}}$$

(4)

2.2.5. Traffic Recognition (TR)

The traffic lights in SVIs determine the TR. TR is essential for meeting the travel needs of the public, influencing the convenience and safety of travel. This study selected traffic signals, sidewalks, and railways as subjects for the TR. The sum of the pixel proportion for these three types of objects represents the TR of the image, calculated using Equation (5):

$$TR={\displaystyle \frac{P_{traffic signal}+P_{sidewalk}+{\mathrm{P}}_{ railway}}{P_{all}}}$$

(5)

2.2.6. Motorization Degree (MD)

The MD is reflected through the traffic at the site, which is determined by the space of roads and the appearance rate of cars. The MD is one of the top priorities in establishing a favorable neighborhood and is a critical element that determines the reachability of the area. Roads, cars, buses, trucks, freight vehicles, small motor vehicles, and railways were selected as the study subjects. The sum of the pixel ratios for these seven types of objects represents the MD of the image, calculated using Equation (6):

$$MD={\displaystyle \frac{P_{ road}+P_{car}+P_{bus}+P_{truck}+P_{freight vehicle}+P_{small motor vehicle}+{\mathrm{P}}_{ railway}}{P_{all}}}$$

(6)

2.3. Quantifying Street Space Quality

To comprehensively evaluate the spatial quality of urban streets, this study introduces the positive index service capability value and the negative index service capability value. The positive index mainly includes four elements: SVI, GVI, PCI, and TR; and the negative indicators include two elements: IEI and MD. Since the trends and amplitudes of evaluation indicators are different, to facilitate indicator comparison and data calculation, the sample data must be normalized and dimensionless [40]. Equation (7) is used to calculate the positive correlation indicators, and Equation (8) is used to calculate negative correlation indicators:

$$y_{ij}={\displaystyle \frac{(x_i-x_{\min })}{(x_{\max }-x_{\min })}}$$

(7)

$$y_{ij}={\displaystyle \frac{(x_{\max }-x_i)}{(x_{\max }-x_{\min })}}$$

(8)

where $x_{\min }$ is the smallest value in the indicator, $x_{\max }$ is the largest value in the indicator, $i$ is the sample number, and $j$ is the indicator serial number.

The entropy weight method (EWM) for calculating weights is a very successful technique for evaluating indicators, which can allocate evaluation indicator weights based on the amount of objective information provided by the entropy value of each indicator, thereby improving the objectivity of the evaluation results [41]. Equation (9) shows the information entropy calculation formula:

$$E_j=-{\displaystyle \frac{1}{\ln n}}{\displaystyle {\sum }_{i=1}^m{y}_{ij}\ln y_{ij}}$$

(9)

where n is the number of indicators, and m is the number of samples.

Equation (10) is used to determine the weight value of each indicator:

$$W_j={\displaystyle \frac{1-E_j}{n-{\displaystyle {\sum }_{j=1}^n{E}_j}}}(j=1,2,\cdot \cdot \cdot ,n)$$

(10)

The calculated weight of each factor is shown in Table 2. Equation (11) is used to calculate the overall value of urban spatial quality.

$$V_j={\displaystyle {\sum }_{j=1}^n{y}_{ij}{W}_j}$$

(11)

3. Results

Table 3. Evaluation results of street spatial quality and six influencing factors.

		Low	Relatively Low	Relatively High	High
SVI (26.15%) 1	Number of sites	817	536	986	5708
	Proportion	10.15%	6.66%	12.25%	70.93%
GVI (12.61%) 2	Number of sites	2150	1836	1380	2681
	Proportion	26.72%	22.82%	17.15%	33.32%
IEI (27.08%) 3	Number of sites	1222	3597	2765	463
	Proportion	15.19%	44.70%	34.36%	5.75%
PCI (0.48%) 4	Number of sites	6833	655	285	274
	Proportion	84.91%	8.14%	3.54%	3.40%
TR (3.90%) 5	Number of Sites	1145	717	636	5549
	Proportion	14.23%	8.91%	7.90%	68.96%
MD (22.72%) 6	Number of Sites	2204	3579	2210	54
	Proportion	27.39%	44.48%	27.46%	0.67%
Street Spatial Quality	Number of Sites	717	1977	3815	1538
	Proportion	8.91%	24.57%	47.41%	19.11%

Note: 1–6 are the average pixel proportions for this factor.

shows the results obtained after evaluating each element and the calculated street space quality according to the indicators shown in Table 1. The longitude of each observation point is labeled as the X coordinate and the latitude as the Y coordinate. The street space quality and the pixel proportion of six influencing factors are labeled as the Z coordinate. ArcGIS was utilized to help visualize our results.

3.1. Evaluation Results Based on Physical Environment and Pedestrian Perception Dimensions

According to the weights shown in Table 2, the scores of the two dimensions of physical environment and pedestrian perception were calculated, and the results are shown in Figure 6. Regarding pedestrian perception, 95.09% of the observation points scored greater than 0.1, while in terms of physical environment, the scores of all observation points were less than 0.1. It reflects that the streets within the Second Ring Road of Changsha have higher sky visibility, higher urban greening, wider streets, and lower motorization, but fewer public facilities and traffic lights. Therefore, the street space quality in the area within the Second Ring Road of Changsha City has significant advantages in terms of pedestrian perception. However, the physical environment needs to be strengthened. The specific areas needing improvement are determined based on six evaluation indicators.

3.2. Evaluation Results Based on Six Factors

According to the evaluation method shown in Table 1, this study divides the pixel proportion of each observation point in various influencing factors into four categories: Low, Relatively Low, Relatively High, and High. The analysis results show that the SVI, GVI, and TR of the streets within the Second Ring Road of Changsha City are high. In contrast, the IEI, PCI, and MD are low. This is consistent with the research results mentioned in Section 3.1.

This study obtained the average pixel ratio of various influencing factors in different administrative regions within the Second Ring Road of Changsha City through statistical analysis. Use the equal-spaced division method to divide it into five levels and conduct visual analysis in ArcGIS software. The results show apparent differences in the quality of street space in different administrative regions regarding GVI and IEI. At the same time, there are minor differences in SVI, PCI, TR, and MD.

3.2.1. SVI

As shown in Table 3, more than 70% of the observation points have a high level of SVI (pixel proportion greater than 20%), and the average SVI of all observation points is 26.15%, which is at a high level. Figure 7a shows that the spatial distribution of observation points under different evaluation levels is relatively even. The average SVI of partial roads in the administrative districts within the Second Ring Road of Changsha City (Figure 7b), from high to low, is 27.6% in Yuelu District, 26.7% in Kaifu District, 24.5% in Furong District, 24.2% in Tianxin District, and 23.5% in Yuhua District. The area east of the river has more observation points at lower levels. It might be because the building density in the area to the east of the river is higher, blocking part of the sky. To a certain extent, this reflects that the area to the east of the river is more economically developed than the area to the west.

3.2.2. GVI

As shown in Table 3, the observed values of the GVI are evenly distributed across various evaluation levels. The highest level (pixel percentage ≥ 15%) is the most frequent, accounting for 33.32%. Overall, the average GVI on the street is 12.61%, which is relatively high. However, observation points at the lower and deficient levels account for 49.54%, roughly equivalent to those at higher levels. This indicates significant room for improvement in the GVI.

Figure 8a shows the spatial distribution of observation points under each evaluation level. In the upper right part of the figure, there are more observation points with relatively lower and lower levels of GVI than those with relatively higher and higher levels. As shown in Figure 8b, the average GVI of roads in the administrative districts within the Second Ring Road of Changsha City, from high to low, is 14.3% in Yuhua District, 13.8% in Tianxin District, 13.0% in Yuelu District, 11.8% in Furong District, and 9.9%, in Kaifu District. The overall difference is significant.

3.2.3. IEI

Table 3 shows that 94.25% of the observation points did not reach a high level of IEI (pixel percentage ≥ 45%). Most observation points are relatively low (44.70%) and relatively high (34.36%). Overall, the average value of IEI is 27.08%, which is relatively low.

Regarding spatial distribution, Figure 9a shows that most observation points with a high IEI are located east of the river. Figure 9b shows that the average IEI of partial roads in the administrative districts within the Second Ring Road of Changsha City, from high to low, is 30.4% in Yuhua District, 28.8% in Furong District, 28.4% in Tianxin District, 26.5% in Yuelu District, and 24.0% in Kaifu District. It also shows a pattern in which the SVI and IEI are inversely related. If SVI is high, the IEI is low, implying streets are more open. Therefore, the conclusion drawn in Section 3.1, which states that the SVI in the area to the east of the river is relatively low, corresponds with the findings in this section that the IEI in the area to the east is relatively high. At the same time, the distribution of SVI and IEI in the entire research area also aligns. Based on the map, the average level of SVI is high, but the IEI remains relatively low.

3.2.4. PCI

As shown in Table 3, the observation points are concentrated at a low level of PCI (pixel percentage ≤ 1%), accounting for 84.91%. The average PCI is 0.48%, which is low. As mapped in Figure 10a, the observation points with a higher PCI are primarily located in the area to the east of the river, suggesting that the level of basic infrastructure in the area to the east of the river area is relatively advanced. The result also reflects that the economy in the area east of the river is more developed. As shown in Figure 10b, the average PCI of partial roads in the administrative districts within the Second Ring Road of Changsha City, from high to low, is 0.54% in Tianxin District, 0.52% in Yuhua District, 0.47% in Furong District, 0.46% in Yuelu District, and 0.39% in Kaifu District; the difference is negligible. From an overall view of the map, the facility provision rate in the entire research area is low, which might cause inconveniences in the daily lives of the people and needs attention.

3.2.5. TR

As shown in Table 3, 68.96% of the observation points fall within the high-level range for TR (pixel percentage ≥ 1.5%). The overall average value for traffic recognizability is 3.90%, which is high. Figure 11a indicates that the observation points for each evaluation level are evenly distributed spatially. Figure 11b shows that the TR average values of partial roads in the administrative districts within the Second Ring Road of Changsha City, from high to low, are 4.42% in Furong District, 4.38% in Yuhua District, 3.96% in Tianxin District, 3.75% in Yuelu District, and 3.47% in Kaifu District. The differences are minor. Overall, the research area has high TR, making it convenient for people to travel. Improvements can be made for areas with lower traffic recognizability by adding road signs, traffic lights, and similar methods. Furthermore, making road signs and traffic lights more artistic and engaging can capture the attention of the people even more.

3.2.6. MD

Table 3 shows that the vast majority (99.33%) of observation points did not reach a high level of MD (pixel percentage ≥ 45%). The overall average value is 22.72%, which is relatively low, suggesting that the research area might generally suffer from traffic congestion issues. From Figure 12a, the area east of the river has more low-level observation points. Given that the area to the east of the river is more economically developed and has a more extensive traffic volume, its traffic congestion issue might be more severe than that of the area east of the river. Figure 12b shows that the average MD of partial roads in the administrative districts within the Second Ring Road of Changsha City, from high to low, is 23.41% in Kaifu District, 23.36% in Yuelu District, 23.31% in Furong District, 21.17% in Tianxin District, and 20.19% in Yuhua District. The differences are minor. In summary, the motorization level of the entire research area is not optimistic.

3.3. Street Spatial Quality

Based on the weights shown in Table 2, the overall street space quality of each measurement point was calculated. Table 3 shows that 66.52% of the observation points have a street space quality at either a relatively high level or a high level. The average street space quality is 12.81%, indicating that the overall street space quality of the research area is good. Furthermore, only 19.11% of the observation points have reached a high level, while 33.48% are at a low or relatively low level, suggesting that the government still needs to focus on urban construction.

Figure 13a shows many low and relatively low observation points near the edge of the study area, such as the southern section of the East Second Ring Road. In addition, the Wujialing area, located on the east bank of the Xiangjiang River, has many low-level observation points that require special attention. Figure 13b shows that the average spatial quality of partial roads and streets in the administrative districts within the Second Ring Road of Changsha City, from high to low, is 18.61% in Yuelu District, 18.47% in Furong District, 18.41% in Yuhua District, 18.19% in Tianxin District, and 17.49% in Kaifu District. The difference is slight, and the overall level is high.

4. Discussion

This study obtained 10,666 SVIs within the Second Ring Road of Changsha City through Baidu Maps. Due to file corruption and image distortion during the transmission process, 8047 valid pictures were obtained. However, this study has the following shortcomings:

• Considering the limited accuracy of the image semantic segmentation tool used, the SVIs acquired from 2019, and some missing images, the precision of the research conclusions may be affected.
• The SVI studied in this paper is not calculated using fisheye images but uses static panoramic street views to align with the most direct perspective of the pedestrian [42]. Because panoramic images might have overlapping pixels, research results can only reflect reality qualitatively.
• It is convenient and efficient to use the image semantic segmentation method to study the quality of street space. However, the environment of street space is relatively complex and is affected by factors such as illumination or occlusion. There will be a certain deviation between the recognition results and the true value.
• The contents of this study have mutual influences, such as the mutual constraints between IEI and SVI and the correlation between MD and TR. Hence, the research results have certain limitations.
• The research scope of this study does not cover the complete administrative area, and the research results can only reflect the street space quality of some spaces in different administrative areas.

Follow-up research can optimize and improve the appeal content and consider more factors that may affect the quality of street space, classify streets according to their functions, such as commercial districts and residential areas, and investigate the spatial quality of different types of streets. Furthermore, we can delve deeper into how these factors interact and their cumulative effect on the overall quality of street spaces. Future studies could concentrate on refining these aspects, thereby enhancing the precision of results. This iterative process of optimization and improvement in our research methods will increase reliability and contribute to a more comprehensive understanding of what makes street spaces attractive and functional. Ultimately, our goal is to provide valuable insights that can guide urban planning and design, creating street spaces that are not only aesthetically pleasing but also practical and user-friendly.

5. Conclusions

This study analyzes static SVIs from Baidu within Second Ring Road in Changsha using a semantic segmentation method based on a deep learning complete convolutional neural network. We obtained a proportion of 150 types of land objects in the images, evaluated them through six selected factors that reflect street spatial quality, and visualized them through ArcGIS software.

The conclusions drawn are as follows:

• Over 70% of the observation points have a high SVI, while the average IEI is 27.09%, which is relatively low. In particular, in the commercial area east of the river within the Second Ring Road, the SVI is low, and the IEI is relatively high, indicating crowded buildings. Since the IEI and SVI are two factors that affect each other, they should be considered collectively to optimize both to a higher level.
• The average GVI is 12.59%, which is relatively low, and the average PCI is 0.48%. This indicates that the greenery within the human field of vision is not sufficient, and the block construction is still subpar. The GVI of Kaifu District is 9.90%, with a PCI of 0.39%, which is the lowest level within the study area. It is suggested that further greenery in street spaces and neighborhood facilities be improved to enhance comfort and convenience.
• The TR is high, with an average of 3.90%, ensuring safety and convenience for the public. And a low MD, with an average of 22.72%. The MD of the five administrative regions ranges from 20.19% to 23.41%, showing minimal differences. This implies potential traffic congestion in the entire study area, necessitating more appropriate road planning.
• In total, 66.52% of observation points have relatively high or high levels of street space quality. Observation points with lower or relatively lower levels are mainly distributed in the southern section of the East Second Ring Road and the Wujialing area on the east bank of the Xiangjiang River. Improvement in various aspects of streets based on the scores in other evaluation indicators within this region is required.
• Overall, the street space quality within the research scope is high, averaging 12.81%. However, only 19.11% of observation points have reached a high quality, and 33.48% still need to reach a higher level. Compared to other areas, the commercial area east of the river has denser buildings, lower levels of greenery, and more congested roads, which necessitates improvements in pedestrian perception. Other areas have significant advantages in terms of pedestrian perception but require enhancements in community public facilities and traffic signage.