Measuring human perception of residential built environment through street view image and deep learning

As an important part of the urban built environment, streets exploring the influence mechanism between the built environment and human perception. It is one of the issues in building healthy cities. In this study, the residential streets of Zhongshan Distict, Dalian were selected as the study site, including Mountain Low-rise Neighborhood, Old Mid-rise Neighborhood, and Modern High-rise Neighborhood. Meanwhile, spatial measurement and human perception perception evaluation of the street environment were based on Deep learning and street view image (SVI). The study used human perceptions as dependent variables, and physical features as the independent variables. Finally, two regression models of positive and negative perceptions were established to analyze the relationship between them. The results showed that in the three types of neighborhood, positive perception was mainly focused on Mountain Low-rise Neighborhood; Negative perception was mainly focused on Old Mid-rise Neighborhood. Greenness, Openness, Natural Landscape, Natural to artificial ratio of the horizontal interface, and Natural to artificial ratio of the vertical interface had a positive influence on positive perception. Pedestrian occurrence rate, Enclosure, and Vehicle Occurrence rate had a negative influence on negative emotive. Greenness was the physical feature that most affected human perception. This study provided a method for objectively evaluating the quality of the street built environment. It is important for promoting the quality of residential streets and public mental health.


Introduction
As the core factor in public psychology, scholars were pay more attention to research human perception and its influencing factors (Wood et al 2017).Previous studies have shown that the urban built environment not only affects human perception, but also related to physical and mental health (Mitchell et al 2013, Yang and Mu 2015, Song et al 2023).Therefore, a good urban built environment can promote the positive human perception (Morris 2003).Currently, How to improve health and environmental problems caused by urbanization are a frontier issue in the research of Human Settlements and Environmental (Shuhua Li 2009, Barahmand et al 2013).
Streets are the carrier for people to perceive the urban built environment, which provide space for social interaction, exercise, and daily life (Wang et al 2022a).The number of residential streets is predominant in cities.Some studies found that negative emotions had a relationship with the deterioration of the street built environment (Galea et al 2005).Therefore, the construction of high-quality residential streets can improve the aesthetics of the streets (Carmona et al 2018), and reduce personal stress (Hagen and Tennøy 2021, Ji et al 2023, Wang and Li 2023).High-quality residential streets have a positive impact on human perception.It led urban planners to build high-quality streets that were people-centered (Shaolan 2007, Olayiwola and Ajala 2022).
In previous studies, scholars mainly studied human perception and mental health from three perspectives: socioeconomic status (SES), social environment, and built environment (Lovasi et al 2009, Astell-Burt et al 2014, Gascon et al 2015).Researches on psychological health include place-related emotional health (Elizabeth et al 2011), mental health (Liu et al 2017), and evaluation of well-being (Steptoe et al 2015, Yue et al 2022b).Living in different street environments, residents have some differences in their emotional states (Bin et al 2017, Liu et al 2022).According to the Dual model of mental health, negative and positive are common dimensions for analyzing human perception (Liu et al 2018).Therefor, different factors in the environmental had the different influence on the two dimensions (Peter andGreenspoon 2001, Keyes andLopez 2002).
For the impact of the urban built environment on mental health, relevant scholars analyzed factors such as urban infrastructure and land use.They pointed out that environmental factors had an impact on human perception (Zeng et al 2023, Villagra et al 2024), such as climate, light, color, and space (Pribram 2002).Since the 1960s, scholars such as Jane Jacobs and Kevin Lynch recognized the importance of the street built environment to human perception and mental health (Lynch 2001, Jane Jacobs 2005).Nasar pointed out that improving natural visibility and environmental amenity could relieve stress.It was helpful for improving human physical and mental health (Nasar 1994, Jenny 2008, Jiang et al 2014, Pfeiffer and Cloutier 2016).In the street spaces, some physical features affect human perception, such as greenery, openness, the height-width ratio of streets, street scale, the size of frontage buildings, and the proportion of tree crown (Jiang et al 2014, Chen 2018, Yang et al 2022a).Meanwhile, previous classical studies shown that these physical features could represent the spatial quality of streets (Arnold 1992, Harvey et al 2015).Environmental stress theory found that some factors had a negative influence on mental health, such as noise, enclosure, and the low proportion of the sky (Roger et al In previous studies, spatial environmental factors were mainly used to explore the influence of the urban built environment on human perception (Huang et al 2023).However, the influencing mechanism between them is complex.Therefore, based on previous studies of urban design quality, this study quantified eight physical features and six perceptual features.We conducted a comprehensive study between street built environment and human perception.The study is as follows: (1) Quantitative measurement urban street space and human perception perception on the large-scale.(2) To explore the linear relationship between the physical features and human perception, and to provide suggestions for people-oriented street construction.(3) Through empirical research, the study will determine the degree to which different physical features impact human perception.
In this study, the residential streets of Zhongshan District in Dalian were selected as the study site.Firstly, 2591 sample points were determined using ArcGIS.By integrating deep learning with Baidu Street View Images (BSVI), we obtained human perception and calculated eight physical features (vehicle occurrence rate, enclosure, greenness, pedestrian occurrence rate, openness, natural landscape, Natural to artificial ratio of the vertical interface, Natural to artificial ratio of the horizontal interface).Finally, the correlation between the street built environment and human perception was explored through correlation regression.The results of this study will provide theoretical support for constructing high-quality residential streets, and help urban planners recognize the relationship between street built environment and human perception.Meanwhile, it contributes to the promotion of public mental health.

Study site and data
Dalian is a significant coastal city of Northeast China, which has been recognized the most livable city in China.Zhongshan District is the part of the urban core, which located on the east of Dalian and share maritime boundaries with Yellow Sea.It has a number of scenic areas, ports and mountains, creating an excellent ecological environment and living conditions.
In this study, we identified the functions of streets in Zhongshan District by using Baidu Maps and the central city land use planning map.The mixed residential, commercial and residential streets were selected as the study sites (figure 1).Based on existing classification methods (Long et al 2019), average number of floors and the construction conditions of the buildings, categorized the study sites as follows: buildings in mountainous areas with an average height of 1-3 floors are classified as Mountain Low-rise Neighborhood, Buildings with an average height of 3-7 floors are classified as Old Mid-rise Neighborhood, Buildings with an average height of more than 7 floors are classified as Modern High-rise Neighborhood.
Figure 2 shows the data collection, processing, and analysis process for this study.
In recent years, urban big data has been increasingly applied to urban research.As a type of urban big data, street view image can accurately reflect the urban street environment.So it was widely used to measure the spatial features of the cities.Baidu Maps provided streetscape data for this study.Firstly, the road network data of the study site was obtained based on the OpenStreetMap (OSM) dataest.Then, it was inputed into ArcGIS to generate a sample point at 50 m intervals along the road and obtained the latitude and longitude coordinates.A total of 2591 sample points were generated.Secondly, Python was used to crawl the street panorama images with the size of 2048 × 624, and 0°heading and pitch angles (figure 3).Meanwhile, the time of street panorama images were set to April to October.It could ensure the seasons are consistent.

Quantitative measurement of street built environment
Semantic segmentation can identify the proportion of landscape elements such as buildings, sky, plants, person, and roads in urban street panorama images (Sun et al 2023).It provided accurate data for urban studies.At present, semantic segmentation models such as FCN, PSPNET, Deeplab, and U-NET have high accuracy and efficiency.In this study, the DeeplabV3+ semantic segmentation architecture was uesd to classify pixel of the images with the ADE20K dataset.It was developed by Google team.DeeplabV3+ is known for its high accuracy, and high efficiency in using smaller training sets.It could solve problem of missing detail information.The 2591 collected panoramic images were inputed into the DeeplabV3+ decoder to obtain landscape element data (figure 4).Eight operable physical features, such as openness, greeness, enclosure, and natural landscape, were selected for study (Ewing and Handy 2009).We took landscape element data into table 1 to calculate the physical features.

Calculating the human perception
The Deep Learning Computer Vision Models was developed by Dubey et al (Dubey et al 2016).It was used to obtain human perception with the 'Place Pulse 2.0' dataset in this study.The six perceptual features in the   residence streets of Zhongshan District (safety, wealthy, lively, beautiful, boring, and depression) were obtained.Some scholars such as Zhang et al and Larkin et al who used this method to measure human perception on the large-scale.Meanwhile, they explored the impact of urban design on public perception (Larkin et al 2021, Zhang et al 2022).
In addition, safety, wealthy, lively, beautiful can represent positive perception (PP).Boring and depression can represent negative perception (NP) (Yu et al 2021).

Statistical analysis
The study used SPSS 26.0 software to conduct Pearson correlation analysis and linear regression analysis.We took positive and negative perception as dependent variables, the relevant physical features screened out were treated as independent variables.Two regression models were established for both positive perception and physical features, and negative perception and physical features.

Spatial distribution of physical features
The study compared the results of physical features in the residential streets of ZhongShan District (table 2).The mean values of the eight physical features ranked as follows: Natural to artificial ratio of the vertical interface (1.860) > Openness (0.471) > Vehicle occurrence rate (0.386) >.
Table 1.The formulas and expressions for the eight physical features.

Physical features Formula Expression
Vehicle occurrence rate C i denotes the proportion of car pixels, T1 i denotes the proportion of truck pixels, B1 i denotes the proportion of bus pixels, T 2 i denotes the proportion of train pixels, the sum indicates the total number of water pixels in each image.Natural to artificial ratio of the vertical interface (2) T3 i denotes the proportion of tree pixels, B2 i denotes the proportion of building pixels in each image. Enclosure (4) T3 i denotes the proportion of trees pixels, G i denotes the proportion of grass pixels, the sum indicates the total number of tree pixels in each image.Pedestrian occurrence rate (5) P2 i denotes the proportion of pedestrian pixels, the sum indicates the total number of pedestrians pixels in each image.Natural to artificial ratio of the horizontal interface G i denotes the proportion of grass pixels, S1 i denotes the proportion of sea pixels, R i denotes the proportion of road pixels, P1 i denotes of pavement pixels.
S2 i denotes the proportion of sky pixels, the sum indicates the total number of sky pixels in each image.Natural Landscape (8) S1 i denotes the proportion of sea pixels, F2 i denotes the proportion of forest pixels.Enclosure (0.325) > Natural to artificial ratio of the horizontal interface (0.065) > Greenness (0.062) > Natural landscape (0.002) > Pedestrian occurrence rate (0.001).In the residential streets of ZhongShan District, the mean value and the standard deviation of Natural to artificial ratio of the vertical interface were the highest.It showed that there were high nature, but the distribution was not balanced with certain fluctuations.The mean value and the standard deviation of pedestrian occurrence rate were the lowest.It showed that pedestrian flow was low, but the distribution was balanced.
By comparative analysis the results of eight physical features in three types of neighborhoods (figure 5), it indicated that Pedestrian occurrence rate, Natural to artificial ratio of the vertical interface, Vehicle occurrence rate were prominent in the Old Mid-rise Neighborhood.In the Modern High-rise Neighborhood, Openness, Natural to artificial ratio of the horizontal interface were prominent.In the Mountain Low-rise Neighborhood, Natural landscape, Natural to artificial ratio of the vertical interface, greenness were prominent.It could be concluded from table 2 and figure 5, in three types of neighborhoods, openness occupied the largest proportion of human sight.It indicated that the residential streets in Zhongshan district had a relatively open sky view and created open street space.The proportion of greenness was low (0.0626), it could be seen that the proportion of green plants in the the human visual range was 6.26%.However, relevant studies showed that the minimum greenness was recognized by human visual range was 15% (Yuan 2012, Gao et al 2020).It indicated the level of greening was poor in the residential streets of Zhongshan District.
From table 2. and figure 6, it could be seen; In the Mountain Low-rise Neighborhood, Natural to artificial ratio of the vertical interface and Greenness were typical physical features.The mean values and standard deviation of Natural to artificial ratio of the vertical interface were the highest.They indicated that the value witnessed a fluctuation.In the vertical interface, the distribution of natural elements were not balanced.The natural elements such as greenery, vegetation, mountain played a dominant role in some resdential streets.They were helpful to enhance the nature of vertical interface.In three types of neighborhoods, the mean value of Greenness was highest in the Mountain Low-rise Neighborhood.They showed the level of greenery was higher, forming a green and healthy environment.
In the Old Mid-rise Neighborhood, Vehicle Occurrence rate, Natural to artificial ratio of the vertical interface and Natural to artificial ratio of the horizontal interface were typical physical features.The standard deviation of Vehicle Occurrence rate was the highest.It indicated that the spatial planning and layout of the Old Mid-rise Neighborhood were complex, the roads were narrow, and the traffic flow was large.The distribution of Vehicle Occurrence rate was not balanced.Meanwhile, the mean values of Natural to artificial ratio of the vertical interface (0.200) and Natural to artificial ratio of the horizontal interface (0.025) were lowest in three types of neighborhoods.It indicated that the buildings, walls and other artificial elements played a dominant role in the visual space, forming a high enclosure in the street space.Greenness was the lowest in three types of neighborhoods.The low Natural Landscape (0.002) and Greenness (0.036) indicated that the nature was poor in the Old Mid-rise Neighborhood.In the Modern High-rise Neighborhood, Openness, Pedestrian occurrence rate and Natural Landscape were typical physical features.The mean value of Openness, Natural to artificial ratio of the vertical interface, Vehicle Occurrence rate were higher.Greenness, Natural Landscape, Pedestrian occurrence rate were lower.The mean value of Openness was the highest, and the standard deviation was medium.It showed that the distribution was balanced.In the Modern High-rise Neighborhood, the most roads were multilane, with a large proportion of street space.They declined the sense of boundary and increased the feeling of spaciousness.The standard deviation of Pedestrian occurrence rate and Natural Landscape were lower, it indicated the distribution of natural elements and pedestrian flow were balanced in the street space.

Spatial distribution of human perception
The study compared the results of human perceptions in three types of neighborhoods (table 3).The mean values of the positive perception ranked as follows: Mountain Low-rise Neighborhood (2.592) > Modern Highrise Neighborhood (2.534) > Old Mid-rise Neighborhood (2.341).It showed that the positive perception was better perceived by the pedestrians in the Mountain Low-rise Neighborhood.The standard deviation of positive perception was the highest.It indicated that the distribution of positive perception was not balanced.In the Modern High-rise Neighborhood, the positive perception was medium, but its standard deviation was the lowest.It showed that the distribution of positive perceptions was balanced.The mean values of the negative perception in three types of neighborhoods ranked as follows: Old Mid-rise Neighborhood (0.827) > Modern High-rise Neighborhood (0.774) > Mountain Low-rise Neighborhood (0.760).It showed that the negatively psychological emotions were better perceived by the pedestrians in the Old Mid-rise Neighborhood.
By the visualization analyzing the results of human perceptions (figure 7), on the residential streets in Zhongshan District, it was found that there was a significant spatial heterogeneity in human perception.Positive and negative perception were distributed in similar spatial patterns.It showed that there was a significant difference between positive and negative spaces in Zhongshan District.
By analyzing the table 3, it could be concluded that the mean value of perceptual features in the residential streets were ranked as follows: Beautiful>Wealthy> Safety> Lively> Boring > Depression.The mean value and the standard deviation of Beautiful were the highest.It indicated that the residential streets in the Zhongshan District were full of aesthetic, but the distribution of Beautiful was not balanced.The mean value of Depression was the lowest and the standard deviation was lower.Due to the Lack of greenery, overcrowding, it indicated that the depression perception were perceived by pedestrians in the overall street space, which was not balanced.
The study compared the mean values of perceptual features in three types of neighborhoods (figure 8).
For the positive perception, the mean values of Wealthy, Safety, Lively and Beautiful in the Mountain Lowrise Neighborhood were higher than the other two types of neighborhoods.In the Mountain Low-rise Neighborhood, the unique buildings and high-quality construction enhanced the aesthetics of the street space.High greenery and natural landscape elements contributed to a tranquil atmosphere and a pleasant environment.They significantly enhanced the livability of the residential streets.In addition, the interface enclosure formed by mountains and street trees also enhanced the safety of the street space.
For the negative perception, the mean values of Boring and Depression in the Old Mid-rise Neighborhood were higher than the other two types of neighborhoods.These factors led to negative perception for the public, such as the slow pace of street renewal.In the Old Mid-rise Neighborhood, the high-density buildings obstructed part of the view and sunlight.Meanwhile, the narrow width of the streets enhanced the sense of enclosure in the street spaces.So Depression was better perceived by public.The buildings along the street have been built for a long time, some building facades were seriously damaged, the composition of the streets was relatively single, lacking landscape elements.These problems were caused the perception of Boring.Mountain Low-rise Neighborhood were characterized by high values of positive perception features, forming positive human perception in these areas.Old Mid-rise Neighborhood were characterized by high values of negative perception features, forming negative human perception in these areas.In addition, the mean value of six perceptual features were medium and poorly characterized in Modern High-rise Neighborhood.

The correlation analysis
The six perceptual features and the eight physical features were conducted by correlation analysis (figure 9).The results showed that Greenness, Openness, Natural Landscape, Natural to artificial ratio of the horizontal interface, Natural to artificial ratio of the vertical interface were positively correlated with positive perception.Vehicle occurrence rate, Enclosure, and Pedestrian occurrence rate were negatively correlated with positive perception.The correlation coefficient of greenness was 0.776.It was strong positively correlated with positive perception.This suggested that the stronger the perception of greenness in the street spaces, the plants were  more helpful for promoting positive perception (Lu et al 2018, Wang et al 2020, Chen et al 2022).The correlation coefficient of openness was 0.461, it was medium positively correlated with positive perception.The results showed that the higher the visibility of the sky in the street spaces, the more positive perceptions were perceived by the public.The correlation coefficients of Natural Landscape, Natural to artificial ratio of the horizontal interface, Natural to artificial ratio of the vertical interface were 0.108, 0.093, and 0.013.They were weak positively correlated with positive perception.It was found that these features were not significantly correlated with positive perception.The correlation coefficient of Enclosure was −0.499. it was medium negatively correlated with positive perception.It was found that buildings and walls obstructed the view and affected public emotion.The correlation coefficients of Pedestrian occurrence rate and Vehicle occurrence rate were −0.161 and −0.312.They were negatively correlated with positive perception.The results showed that many pedestrians and vehicles could cause cluttered street spaces, which led to negative perception.

The regression analysis
In the study, two regression models were used to establish to between physical features and perceptual features.According to the classification of human perception, positive and negative perceptions were individually used as dependent variables for the two regression models.The physical features screened by the correlation analysis were used as independent variables.The stepwise regression analysis was used to allow the independent variables to enter the models.The unsuitable physical features were removed one by one.Finally, two regression models were established (table 4).The R 2 values of the two models were 0.882 and 0.863.It indicated that the regression lines fit the observations well.Table 3 showed that the significance P-values of the F-tests were 0.000 in the two models.Their Sig.values were less than 0.05.The VIF values of the independent variables in the two equations were less than 10.Therefore, it showed that human perceptions established two valid regression analysis models with physical features.

Discussion
Through the study of the residential streets in Zhongshan District, Dalian, we explored the relationship between human perception and the street built environment.By regression analysis, the two regression models were established based on positive and negative perceptions.The results showed that the street built environment was able to affect human perception.It was similar to the findings of (Poortinga et al 2007).According to the comprehensive analysis of positive perception and physical features, the study found that the natural elements had a positive impact on positive perception, such as plants, sky visibility, mountains, and sea (Richardson et al 2013).Mountain Low-rise Neighborhood with high nature and high openness could provide positive perception to the public (de Vries et al 2013, Zhang et al 2018a).Based on a comprehensive analysis of negative perceptions and physical features, the high enclosure had a negative impact on human perception.It was formed by less sky visibility, lush plants, and high-rise buildings (Li et al 2021).In addition, the large traffic flow caused by pedestrians and vehicles had an impact on negative perception.

The influence of street built environment on human positive perception
In this study, the results of plants and public positive perception were consistent with the results of previous studies (Ambrey and Fleming 2013, Tsurumi and Managi 2015, Fleming et al 2016, Zhang et al 2017).Plants in the street spaces were considered to be a positive factor for promoting physical activity and mental health (Zhu et al 2023).A good green space can improve the human perception of beautiful and reduce perception of depression and boring (Subiza-Pérez et al 2020).Lush vegetation and diverse plant communities can improve the ecology of the street space, block noise.Meanwhile, they are helpful for enhancing the amenity and comfort of the street space, and creating a livable street environment (Watts 2017, Wood et al 2017, Phillips et al 2023).Green spaces offered residents access to nature, and relieve psychological and mental stress (van den Berg et al 2010, Jiang et al 2014, Ta et al 2021).People who live in residential streets with high levels of greenery had better mental health than live in low levels of greenery (Sugiyama et al 2008, Francis et al 2012).Urban greenery can increase aesthetic and attractiveness.Furthermore, it is a positive influence on human perception and psychology.Previous studies shown that green space is negatively correlated with negative perceptions such as anxiety (Maas et al 2009, Losert et al 2012, de Vries et al 2016).In addition, living in the residential streets such as green space and scenic spots can help relieve psychological distress and regulate emotions, especially good for mental health (Kelly et al 2018, Kim et al 2023).
Openness was one of the important physical features affecting the comfort of street space.The sky element could improve human perception and psychology (Adebara et al 2022, Xu et al 2023).In the street space, the high openness can increase the aesthetic of the streetscape, thereby promoting the attractiveness and comfort of the street built environment (Ma et al 2021).The results of previous studies were similar to our study.For example, Tang and Long found that the openness could increase the attractiveness of streetscapes (Tang and Long 2019).
The open view of the street space positively influenced the happiness and vitality of the residents (Zhang andLin 2011, Zhang et al 2018b).In the Zhongshan District, Old Mid-rise Neighborhood can increase the visibility of the sky by removing some old buildings and widen the sidewalk.It is beneficial to increase the public positive perceptions such as beautiful, wealthy and lively (Dai et al 2021).
In addition, blue space was one of the physical features influencing positive perception (Nagata et al 2020).The human perception of blue spaces could help suppress negative perceptions (Wei et al 2022).The rich natural atmosphere can enhance the human perception of aesthetic and improve their emotion (Lyu et al 2022).By increasing the visibility of sea and mountain, there are helpful to provide the public with positive perception (Milligan and Bingley 2007, Nielsen and Hansen 2007, Gillis and Gatersleben 2015, Hung and Chang 2022).Therefore, urban planners can combine green and blue spaces with urban development.It plays a important role in promoting mental health.A large number of experimental results are consistent with our study, the natural environment is more effective than the artificial environment in reducing stress (Maas et al 2009, Triguero-Mas et al 2015, van den Berg et al 2015, Dadvand et al 2016).For example, by combing greenery with building façade, the proportion of artificial elements is reduced in the street space.It is helpful to attract the human attention, and reduce the human perception of negative perceptions such as boring (Wang et al 2022a).

The influence of street built environment on human negative perception
In the street space, overcrowded buildings not only obstruct the view, but also increase the traffic pressure.It led to the perception of anxiety and depression, which could affect the physical and mental health (Nordbø et al 2018, Ji et al 2023).In addition, the high enclosure caused by high-rise buildings.It had a negative impact on psychological recovery, The higher enclosure had the larger the influence on the human perception (Lindal and Hartig 2013).Baum et al pointed out the relationship between crowded streets and psychological distress (Cohen et al 1986, Baum andPaulus 1987).The denser the buildings and the less proportion of the sky could the stronger enclosure in the street space.The more psychological distress was exhibited by residents walking in high-density residential streets, which compared to those in low-density residential streets (Maxwell 1996, Kanyepe 2023).Previous studies pointed out that reducing the sense of oppression caused by buildings and greenery were helpful for decreasing the negative impact on human perception in the high-rise residential streets (Wang et al 2021).Moreover, the quality of buildings in the residential streets was directly proportional to psychological stress, The damaged facades and old buildings are negatively correlated with aesthetic of streets.Low-quality street environment brought insecurity (Maxwell 1996, Hamim andUkkusuri 2023).
Moving vehicles exacerbated the danger of the road, and brought the insecurity to the public in the street space.At the same time, noise from vehicles had a negative impact on the mental health of the public (Ma et al 2020).Disorderly parked vehicles on both sides of the street increased the public negative perception.With the increase of crowd density, the negative effect on the human positive perceptions was larger (Zhao and Li 2021).Streets provide more places for activities and reduce the crowd density in the space, which is conducive to expressing more positive perceptions.

Limitation
This study quantified the physical features of streets and human perceptions on the large-scale.At the same time, the study explored which physical features affect psychological emotions.However, The limitations of the study are also worth discussing.
Street view images were obtained from a single time dimension.In the future, we can break the time limitation and collect multi-time street view images.It is helpful to explore the evolution of the street built environment.In addition, multi-season street view images can be obtained to help researchers better understand the impact the built environment on human perception in different seasons.
The study only considered the relationship between built environment and human perception.Future research can be used to measure physiological responses by the physiological monitoring technologies such as EEGs, eye movement tracking, and other techniques.Then, the visual physiological indices of the public will be obtained as mediating variables.In addition, future studies can consider other factors (such as attention restoration and physiological stress recovery) to further analyze the influence mechanism between human perception and built environment (Markevych et al 2017, Dzhambov et al 2018, Meng et al 2023).
In addition, the evaluation of human perception can be further refined in the future studies.Although this study has enriched the studies on visual perception.The perception of sound and smell should be considered in the future studies.In addition, the culture, history, and activities of the built environment are difficult to represent through images.By using POI, sound field measurement technology, the studies in the future will conduct to more comprehensively analyze the influence mechanism of urban built environment and human perception (Yue et al 2022a).
Based on computer vision techniques and big data, the study measured the street built environment and human perception.In the future, we could further quantitative analysis of urban planning and public health.It will contribute to more comprehensively analyze the influence mechanism of social environment, economic vitality, public health and built environment.This scientific and precise conclusions will be more valuable for both academic researchers and urban planning practitioners.

Conclusion
The residential streets of Zhongshan District in Dalian were selected as the study site.By combing machine learning and street view image, the study was conducted to quantitative measurement of street built environment on the large-scale.Meanwhile, the study evaluated six perceptual features, such as safety, wealthy, lively, beautiful, boring, and depression.Then, the two regression models of human perception and street physical features were established from the positive and negative perceptions.
From the physical features, the results showed that the proportion of Natural to artificial ratio of the vertical interface and Openness were mainly physical features in the residential streets of Zhongshan District, Greenness occupied the lower proportion.From the distribution of human perception, the positive perceptions such as Safety, Wealthy, Lively, Beautiful were mainly focused on the Mountain Low-rise Neighborhood.The negative perceptions such as Boring, and depression were mainly focused on the Old Mid-rise Neighborhood.Greenness, Openness, Natural Landscape and Natural to artificial ratio of the vertical interface had the positively influence on positive perception.In the residential streets, natural elements played a important role in promoting public positive perception and improving mental health.
The study explored the relationship between street built environment and human perceptions.It revealed the physical features in the street built environment that affect human perception and mental health.It not only provides the theoretical and technical support for academic researchers and urban planning practitioners, but also is helpful for promoting mental health, human settlements and environment.
1991, Lin et al 2023).The traditional methods for quantifying human perception included questionnaire surveys (Sampson et al 1997, Volker et al 2006, Tennant et al 2007, Furukawa et al 2008), photography, emoji-based sentiment classifications (Fernández-Gavilanes et al 2021, Ahanin and Ismail 2022), cognitive maps (Chen 2018), and wearable sensors (Sv and Ittamalla 2021, Tsou et al 2021, Wang and Li 2023).The development of deep learning and social media provided new opportunities to obtain the human perception within the urban built environment (He et al 2023a).WeiBo data is characterized by real-time update and massive data.It provides a basis for researching human perception by combining the text sentiment analysis module in deep learning (Yang et al 2019, Yang et al 2022b, Li and Ahn 2024).Quarcia pointed out that crowdsourcing effort could achieve quantitative human perception on the large-scale (Quercia et al 2014, Middel et al 2019).Place Pulse 2.0 was a data that made by Massachusetts Institute of Technology (MIT).It included street View images of 56 cities. Dubey et al trained the Place Pulse 2.0 dataset to develop a deep-learning computer vision model.The model has a 74% accuracy in predicting perception such as safety, wealthy, lively, beautiful, boring, and depression (Dubey et al 2016, Larkin et al 2021).So it is used to obtain human perception of the urban built environment at a large scale (Yao et al 2019).Other scholars conducted many studies about human perception by these perception features (Wu et al 2023, He et al 2023b).Some scholars studied the street built environment from a quantitative perspective.The traditional methods for researching mainly included expert evaluation (Ewing and Cervero 2001, Sun et al 2021), field survey (Xu and Kang 2014), photography (Hao et al 2016), manual extraction.Currently, some open data is used in urban studies, such as street view image (Zhang and Hu 2022), POI big data (Wu et al 2021, Pan et al 2023), GIS.They made researches more refined and data acquisition easier (Long and Zhou 2016).Relevant studies shown that street view image could represent the objective environment in the city (Tao et al 2022; Aikoh et al 2023).At present, it is widely used in the researches such as mental health (Wang et al 2022b, Yuan et al 2023), perceptual evaluation (Lyu et al 2022), walkability (Li et al 2022, Jeon and Woo 2023), street visual quality, and 3D city reconstruction (Lu et al 2024).With the development of computer vision technique, semantic segmentation improved the high efficiency of processing massive data.More and more researches were combining semantic segmentation with street view image (Gonzalez et al 2020, Nagata et al 2020), which were widely used in urban planning and landscape researches (Wang et al 2022a).Landscape elements in the street view image were collected by convolutional neural network, such as sky, streets, buildings, plants.Scholars confirmed the validity by comparing the results of manual extraction and semantic segmentation (Aikoh et al 2023).

Figure 5 .
Figure 5. Physical features in three types of neighborhoods.

Figure 6 .
Figure 6.The spatial distribution of physical features in three types of neighborhoods.

Figure 7 .
Figure 7.The spatial distribution of human perceptions in three types of nighborhoods.

Figure 8 .
Figure 8.The perceptual features in three types of neighborhoods.

Figure 9 .
Figure 9.The correlation analysis of the human perceptions and physical features.

Table 2 .
The result of physical features for various type neighborhoods.

Table 3 .
The result of perceptual features for various type neighborhoods.