Using google street view panoramas to investigate the influence of urban coastal street environment on visual walkability

Urban street walkability can effectively promote public health and the construction of livable cities. In addition, the coastal streets play a positive role in showing urban vitality and image. Due to the growing leisure needs of residents, measuring the visual walkability perception (VIWP) in urban streets and exploring the influence mechanisms of urban coastal street environments on VIWP have theoretical and practical significance. However, the methods of the previous walkability studies have limitations in terms of cost, time and measurement scale. Based on Google Street View Panoramic (GSVP) image data, this study used the semantic difference (SD) method with virtual reality (VR) technology to evaluate the VIWP of Fukuoka coastal streets. Meanwhile, the proportion of streetscape elements was extracted from GSVP images by semantic segmentation. The correlation and regression analyses were performed between the VIWP evaluation values and streetscape elements. Then, the regression model of the VIWP and the streetscape elements was established. The results showed that the natural features had a positive influence on VIWP in coastal streets. Correspondingly, trees were the strongest contribution rate for the VIWP, followed by shrubs, grasses and water, however, buildings and cars had a negative influence on VIWP. The method extends previous studies for measuring walkability, and optimization strategies were proposed to improve the visual quality of the coastal streets. It can be applied in the construction and management of walkable coastal street environments.


Introduction
With global urbanization and the popularity of car-dependent cities (Wang and Zhou 2017), walkable street environments are becoming increasingly scarce (Balsas 2017). As a 'green' transportation mode, walking not only is environmentally friendly but also beneficial for residents physical and mental health (World Health Organization 2010). Walking behaviors play a vital role in promoting active life and urban vitality (Cerin et al 2007, Gauvin et al 2008, Sung et al 2015. Related studies have shown that urban streets with good walkability promote the social and economic prosperity of neighborhoods (Duncan et al 2012).
Urban coastal streets as an important part of urban space (Liu 2016), it has rich and varied landscapes, including ecological landscapes, artificial landscapes, and cultural landscapes (Gan 1998). Urban coastal streets play a positive role in presenting the urban style and features, promoting urban vitality (Ewing et al 2006, Lynch 2001, Shach-Pinsly et al 2021. Furthermore, the quality of waterfront space impacts public behaviors , the physical and mental health of residents (Foley and Kistemann 2015, Kati and Jari 2016, Othman et al 2021. Several studies have explored the impact of urban waterfront spaces on public aesthetics and emotions (Yang and Li 2013, Ragheb andEL-Ashmawy 2020, Sun 2021). The design and construction of pedestrian-friendly coastal streets have attached widespread attention.
However, Walkability is a subjective concept that describes the quality of the environment, walking behaviors and decisions are influenced by individual differences, and walkability depends on human perception (Wang and Yang 2019). In previous studies, subjective measure methods of walkability usually include interviews and observing walking activities (Saadi et al 2022), they have significant limitations in terms of cost, time and measurement scale. With the development of mapping services, computer-aided auditing methods based on street view images are becoming increasingly popular for measuring walkability (Blečić et al 2018, Wang and Yang 2019, Ki and Lee 2021. Vision is the main way for humans to perceive the environment, so the visual walkability and visual walkability index were proposed as indicators to measure the walking environment of streets (Zhou et al 2019a). However, the visual walkability indices constructed from different street physical features are hardly representative of the subjective perception of visual walkability.
Therefore, applying an effective method to quantify the VIWP of coastal streets is the premise and the key to build urban coastal streets with walkability. In addition, exploring the complex relationship between the coastal street environment and human VIWP is necessary to create a coastal street environment conducive to physical and mental health. The critical problems in this study are: (1) The methods of evaluating the VIWP and measuring the streetscape elements in coastal streets efficiently and accurately.
(2) The spatial heterogeneity of VIWP in Fukuoka coastal streets.
(3) The relationship between VIWP and streetscape elements.

Literature review
Walkability was defined as the extent to which the street environment and form support and encourage walking (Southworth 2005), the better service facility, higher landscape environment quality and convenient travel paths improved the street walkability Abley and Hill (2005). Previous research has explored the relationship between walkability and the variables of built environment characteristics at a macro level (Moura et al 2017), including the 5Ds framework of density, diversity, design, destination accessibility, and travel distance (Kang 2018. Related Studies have explored that when people live in higher connected neighborhoods, they are more likely to have access to main roads and higher walking potential . The short distances traveled can effectively encourage people to go out and walk (Kowaleski-jones et al 2018). However, these widely used variables were insufficient to capture the effects on the pedestrian walking experience at the street level and micro level (Lynch 2001). Ewing and Handy (2009) pointed out that the physical features of streets were created by their landscape elements, they influenced the quality of the walking environment direct or indirect through people's perceptions Ewing and Handy (2009). The street greenery and water can not only provide a series of health and well-being benefits (Hartig et al 2014, Frumkin et al 2017a, Bratman et al 2019 but also influence pedestrians' thermal comfort, prompting pedestrians to walk under the shade (Jiang et al 2014). The studies of visual evaluation indicated that enclosure was an important factor for human responses to the environment. The impact of street enclosure on human perception has been repeatedly documented Long 2019, Dai et al 2021). The open street enhances the impression of safety (Montgomery 2014), it potentially encourages walking activities for residents (Arnold 1980, Cauwenberf et al (2012). In addition, the movement factors in the street, such as motorized vehicles and non-motorized vehicles, can affect pedestrians' sense of safety and comfort in the street space, thus affecting the pedestrian walking experience (Li et al 2022a(Li et al , 2022b. Related studies have gradually incorporated micro-scale and subjective-level walkability criteria into walkability evaluations (Bivina andParida 2019, Howell et al 2019). Tarek et alcombined macro and micro design indicators of the building environment to investigate the walkability of Cairo's streets Tarek et al (2021).
Street walking quality usually is responded to the behavior, walking experience and feelings of pedestrians in the street space (Sallis et al 1997, Ewing andHandy 2009), so walkability is a subjective concept. People's walking behavior is generally influenced by visual and non-visual perceptions of humans. In recent years, related studies have proposed visual walkability indices to evaluate visual walkability. Li et al (2022aLi et al ( , 2022b described visual walkability perception as whether the built environment of a street visually encourages people to walk, emphasizing the influence of visual elements on walk willingness Li et al (2022aLi et al ( , 2022b. Although many studies have been conducted in this field, there are no standard methods to evaluate walkability. In previous studies, relevant researchers measure urban street walkability using participatory techniques such as field reviews and questionnaires (Saadi et al 2022). Although these methods have the advantage of being close to reality and analyzing individual differences, there are considerable limitations in terms of cost control, time consumption and scale of measurement. In addition, the assessment of walkability is based on an index system, such as the Pedestrian Environmental Review System (PERS,n.d.), the Neighborhood Environment Walkability Scale (NEWS, n.d.). These methods have a detailed understanding of urban street environment through multi-level evaluation indices (Campisi et al 2021), however, some important but not easily quantifiable features may be overlooked (Zhou et al 2019a).
With the development of sensing technologies and map services, related map service companies (Google Maps, Baidu Maps) have collected a wealth of street view images with GPS data from all over the world (Helbich et al 2019, Rzotkiewicz et al 2018. These freely available image data have become the main data source to explore the relationship between the urban environment and human perception . For example, identifying street composition variation. (Tang and Long 2019), assessing the openness of the sky , measuring the street greenery and physical activity (Lu 2019, Ye et al 2019) and visually evaluating neighborhood walkability (Zhou et al 2019a). Although it provided a view similar to the human eye, but it was difficult for them to provide a real on-site walkability perception (Kim and Lee 2022).
VR technologies and Deep Convolutional Neural Network (DCNN)-based deep learning algorithms overcome the limitations of previous methods and show great potential for processing large-scale data (Li et al 2015, Zhang et al 2018, Qiu et al 2021. The VR technologies with panoramic images provide a near-realistic street environment for evaluators (Bellazzi et al 2022), some researchers examined the streetscape environment by showing street panoramic images through VR devices (Mouratidis and Hassan (2020)). Li et al measured visual walkability perception by using the VR panoramic-based deep learning framework, and the validity of the visual walkability perception scores were verified by an on-site audit . The VR technologies are helpful for the evaluator to capture the street environment information in detail, making subjective perception judgment close to the real environment (Kim and Lee 2022). Earlier methods of semantic segmentation were mainly based on the color of the pixels in images, but it has an obvious disadvantage when two different objects are the same or similar color (He and Li 2019). In recent years, with the latest development of deep learning algorithms, the semantic segmentation framework, such as FCN, ResNet and SegNet, can process the visual information in images using a deep convolution neural network, which can be used to effectively identify streetscape elements in street view images, such as building, sky, plant, road, and sidewalk et al. It is helpful to establish a solid foundation for more reliable research on higher-quality urban streets and human perception , Wu et al 2019, He and Li 2021.
In this study, we chose the Fukuoka coastal streets as the study area. The study used SD method and VR technology to evaluate the VIWP of the coastal streets. The semantic segmentation method was used to extract the proportion of coastal streetscape elements. The study explores the complex relationship between the coastal streetscape elements and the human VIWP. This work has practical implications for the construction of walkable coastal street environments.

Research framework
The research framework mainly includes three phases (figure 1). First, the study used Open Street Map (OSM) to obtain street network data in the study area and used the 'create random point' function of ArcGIS 10.2 to create GSVP image collection points. GSVP images were automatically obtained by writing a Python program combine with the Google Street View Map Application Programming Interface (API). Second, the study used the SD method combine with VR technologies to evaluate the VIWP of each GSVP image. To obtain the objective and accurate proportion of streetscape elements, this study extracted streetscape elements from each GSVP image through Semantic Segmentation. Thirdly, we used ArcGIS software to visualize the VIWP values of coastal streets and analyze their spatial heterogeneity. The study explored the influence mechanism of different streetscape elements on VIWP through correlation analysis and regression analysis. Finally, the regression model of the VIWP and streetscape elements was established.

Study area
Fukuoka is an important port city in Japan. It is located along the coast of Hakata Bay in the northern part of Kyushu island (figure 2). Fukuoka City has a resident population of 1.53 million, and it is the most populous city on Kyushu Island. In addition, as a typical coastal and tourist city in Japan, Fukuoka is near the mountain and sea, with the pleasant features of the natural environment. The natural environment organically combines with the artificial landscape, which forms a unique urban landscape. The study site is located on the northern waterfront of Fukuoka, and the total length is about 26.5 kilometers. The western coastal streets contain Shima Road and Marina Road. The middle coastal streets include Yokatopia Road, Kuromongawa Road and the west part of Nanotu Road. Eastern coastal streets includes the east part of Nanotu Road and National Highway No.3. Fukuoka coastal streets cover the main waterfront area of Hakata Bay, and it is an important area for tourism and recreation. Therefore, Fukuoka coastal street is a suitable research object to explore the relationship between the coastal street environment and pedestrian VIWP.

GSVP images collection
Using street view images to analyze urban environment has became increasingly common in urban science research (Cheng et al 2017). In order to comprehensively display the Fukuoka coastal street environments, this study used Arc GIS10.2 to set GSVP sample points at intervals of 50 m along the road network (Ye 2018). In addition, the latitude and longitude of each sample point were collected. A total of 531 sample points were generated in the Fukuoka coastal streets. The pitch angle of the GSVP image was set as 0• to achieve a more realistic view of pedestrians (Qiu et al 2022). The study collected the GSVP image of each sample point by writing a Python program combine with the Google Street View Map API. The resolution of the GSVP images was set to 2048 × 624. In addition, this study controlled the influence of seasonal changes on streetscape by checking the time when street view images were taken. Figure 3 shows an example of a downloaded GSVP image in Fukuoka coastal streets.

The evaluation of VIWP
In this study, the VIWP of coastal streets was evaluated by combining the SD method with VR technology. The SD method, also known as the semantic differential method, was proposed by Osgood in 1957(Osgood et al 1957. The SD method has been widely used in environment evaluation studies due to its better applicability (Sun et al 2021, Lyu et al 2022. The evaluation process of SD method usually involoves 20 to 50 observers who need a certain amount of professional knowledge. Therefore, the observes included 20 master students and 15 professional teachers who had academic backgrounds in architecture, urban planning, and landscape design (table 1).
VR devices have a more realistic live experience compared with the traditional slides medium, so the study used the VR devices as a medium to evaluate VIWP. During the evaluation process, the adjective pairs of VIWP (unfit for walking and fit for walking) were first explained to the 35 observers, and they were asked to pay attention to the suitability of the scenes in the sample photo for walking rather than the photo quality. Every participant was asked to independently evaluate the GSVP images which were randomly played. Each GSVP image was shown 20 s, evaluation criteria on a five-point scale (1-5), and the positive and negative adjectival expressions were compared with the current event. In this case, the higher score was closer to the meaning of the adjectives on the right. 35 questionnaires were returned, after checking the evaluation results, all of them were valid.

Streetscape elements segmentation and extraction
The research used the semantic segmentation method based on the codec structure SegNet to identify the proportion of streetscape elements in GSVP images. SegNet is an open-source image segmentation project released by the University of Cambridge in 2015. This deep learning-based approach can recognize and classify the semantics of various streetscape elements in images, at the same time, its accuracy can reach the pixel level (Badrinarayanan et al 2017). SegNet performs better compared with other semantic segmentation architectures when extracting spatial features from low-resolution images (Wang et al 2022). In addition, the study used ADE 20 K dataset as the training dataset. The ADE 20 K dataset contains 150 objects from daily life that can adequately capture the complexity of real-world urban scenes. We input the collected GSVP images to SegNet and classify the streetscape elements into color categories by SegNet decoder, then, obtained the proportion of each streetscape element in the GSVP images. As shown in figure 4, which can clearly extract the key streetscape elements from the GSVP images.

Statistical analysis
The VIWP values and streetscape elements were subjected to Pearson correlation analysis by using SPSS 25.0 software. The streetscape elements with significant correlation were identified and retained. The VIWP values were accepted as dependent variables and the regression model of the VIWP and the streetscape elements was established using linear regression analysis.

Fukuoka coastal streets VIWP analysis
The VIWP values of coastal streets were evaluated by using the SD method combining with VR devices. We collected and normalized the VIWP values for all sample points in Fukuoka coastal streets (Appendix A). The top 177 samples were classified as high VIWP samples, the bottom 177 samples were classified as low VIWP samples, and the other 177 samples were classified as moderate VIWP samples. The figure 5 represented the spatial distribution of VIWP in Fukuoka coastal streets. The spatial distribution of VIWP in Fukuoka coastal streets was mainly influenced by the street function. High VIWP samples are mainly distributed on the Western coastal streets, especially in the east part of Shima Road and the west part of Marina Road. In addition, some high VIWP samples were distributed on the Middle coastal streets. There are residential houses with small volume. The low-rise buildings were scattered, and the lush vegetation was on both sides of the street. Low VIWP samples were mainly distributed on Nanotu Road and National Highway No.3, These roads were located in the central business district of Fukuoka, the function of the streets were commercial. The street had a heavily artificial landscape and a lot of flow density of motorized vehicles. The tall and dense buildings on both sides of the street. In addition, the street space lacks greenery.
By analyzing the VIWP mean values of three different areas in Fukuoka coastal streets (table 2) the result showed that: Western coastal street VIWP (0.224) > Middle coastal street VIWP (0.054) > Eastern coastal street VIWP (−0.5624). The Western coastal street had the highest VIWP, the VIWP of the middle coastal street was moderate, and the Eastern coastal street had the lowest VIWP.
By comparing the distribution of high and low VIWP samples in different areas (table 1), it could be seen that the VIWP of the Western coastal street was outstanding (0.224), the proportion of high VIWP samples in the Western coastal street is the highest (46.72%). The Western coastal street was located on the edge of Fukuoka city center, and near the coastline. Low-rise residential buildings on both sides of the streets. The street spaces contained several ecological parks, the overall street has excellent greenery with a strong natural atmosphere. However, the standard deviation of the Western coastal street was highest (0.93), it indicated that the Western coastal street VIWP was an uneven distribution. In the Western coastal street, high VIWP samples were mainly distributed on the east part of the Shima Road (62-158, 174-204), these roads were near Nagatare Park, Ikinomatsubarakaigan Forest Park and Odo Park, at the same time, the streets were close to the coastline and have unique natural landscape environments. The low VIWP samples mainly distributed on the west part of Shima Road. There were a lot of residential buildings on both sides of these roads, the high proportion of the artificial landscape in street space.
The VIWP of the Eastern coastal street was the worst in Fukuoka (−0.562). The proportion of low VIWP samples in the Eastern coastal street is the highest (60.02%). The Eastern coastal street located in the Hakata district and East district in Fukuoka city, they were close to the commercial center of Fukuoka. The street region had numerous artificial landscapes, with densely packed commercial buildings on either side of the streets. The landscape elements in some street spaces were messy. In addition, the street roadway has a higher density flow of vehicles.
As shown in figure 7, the low VIWP samples generally showed low coverage of street greenery and a high proportion of buildings (481-498). The buildings on both sides of the street were dense and large volume, it formed an enclosed street space (360-372, 388-406, 521-532). The interface of buildings was chaotic in street. The streetscape was cluttered by a large number of disorderly street furniture and signs (1-6, 34-61, 159-165, 338-344). The artificial landscapes were monotony and they lacked diversity (446-463). There were many vehicles on the roads, the streets showed a crowded scene (423-442, 470-476).

Streetscape elements and VIWP in coastal streets
In order to explore the influence of the coastal street environments on VIWP, the study used semantic segmentation combining with ADE 20 K dataset to extract the proportion of streetscape elements from GSVP images. Table 3 presents the proportion of the top 10 streetscape elements extracted from GSVP images of  coastal streets in Fukuoka. These include the sky, road, building, tree, sidewalk, shrub, car, fence, grass and water.
We ensured that the linear relationship, the normal distribution, the equal variance and the Independence assumption were satisfied between the variables before calculating the Pearson correlation coefficient. And then, The SPSS 25.0 software was used for correlation analysis between VIWP values and 10 streetscape elements of all samples, the streetscape elements with correlation were retained. The VIWP values were taken as the dependent variables and the retained streetscape elements were taken as the independent variables to conduct stepwise regression analysis for creating the regression model of the VIWP and streetscape elements for coastal streets.
The correlation analysis of VIWP values and 10 streetscape elements of the coastal streets showed that (table 4) the p-values of sidewalk and fence were more than the significance level (significance level of a = 0.05), indicated that there were no correlation significant with VIWP values, they were removed. Whereas other streetscape elements were correlation significant with VIWP values. The tree, shrub, grass, and water were positively correlated with VIWP, which the tree has the strongest correlation (0.793) with VIWP, followed by grass (0.543) and shrubs (0.298). The result indicated that street greenery can significantly enhance the VIWP for pedestrians in coastal streets. The building (−0.567), road (−0.320), car (−0.305) and sky (−0.178), were negative correlation with VIWP. it showed that an increased proportion of buildings in human view has a strong negative impact on VIWP in the coastal streets.
In order to establish the regression model of VIWP and streetscape elements. The VIWP values were taken as the dependent variables and the retained 8 streetscape elements were taken as the independent variables to conduct stepwise regression analysis. Stepwise model building is a statistical method that is commonly used in regression analysis to select a subset of relevant predictor variables for inclusion in a model. In this study, we chose the forward selection for stepwise model building and determined by a pre-defined significance level (pvalue less than 0.05). The F significance level of sky and road was founded to be more than 0.05 during the F-test, the result indicated that the sky and road were less significant to the regression model, so the sky and road were excluded. The related 6 landscape elements and VIWP values were re-run for regression analysis. As shown in figure 8, the regression standardized residual showed that the standardized residuals basically followed a normal distribution, which meaning the residuals basically obeyed the normal distribution and the result indicated that the model assumptions were satisfied for these data. Table 5 showed that the adjusted R 2 was 0.746 in stepwise regression model 6, the regression analysis was statistically significant. In addition, the significance value for the F-test was less than 0.05, which showed that the independent variables were statistically significant predictors of the dependent variable. The significance value of the T-test is less than 0.05, the regression coefficients passed the significance test. the overall regression relationship is significant. The t-value significance level is 0, it indicated that the independent variables were significant in these models. The VIF values of the independent variables were less than 10, it indicated that independent variables were no multicollinearity between streetscape elements. All the above tests indicated that the results of the regression model of VIWP and streetscape elements were valid.
By table 5, it could be seen that there are 6 streetscape elements influence the VIWP in the coastal streets. The standardized Beta coefficients showed that the streetscape elements influence the VIWP in the order of tree, building, shrub, car, grass, and water. Among them, the tree, shrub, grass, and water have a positive contribution to VIWP. The car and building have a negative contribution to VIWP in the coastal streets.

Discussion
In recent years, street walkability has been widely discussed, however, few researches have discussed the people VIWP and the impact of coastal streetscape elements on VIWP at the micro level. In this study, we used the SD method and VR technology to evaluate the coastal street VIWP based on the GSVP. As a new method, the method reduces the difference between auditing the real-world based on the street view images and on-site assessment, and it can be extended to other studies about subjective perception based on the vision.
In addition, the study used the semantic segmentation method to extract 10 main coastal streetscape elements from GSVP images. The VIWP values were used as the dependent variables and the retained streetscape elements were used as independent variables, multiple linear regression analysis was conducted. This study systematically explored the influence of the streetscape elements on the VIWP in the coastal streets. The results can help urban planners and designers to understand the main landscape elements that influence the VIWP of coastal streets. It helps to create a walkable coastal street environment.

The influence of natural features on VIWP
The regression results showed that the trees had the largest contribution rate on VIWP. The other variables about nature features (shrubs, grasses, and water) also had a significant positive contribution to the VIWP. This result indicated that green streets are more friendly for pedestrians walking. Calogiuri and Elliott (2017) also noted that greenery can improve the pedestrian walking experience in the streets by increasing shade (Hahm et al 2017) and reducing stress (Wang et al 2020a(Wang et al , 2020b. Urban streets greenery not only has the function of purifying air Chen and Jim (2008), mitigating the urban heat island effect (Ferla 2020) and reducing noise pollution (Nourmohammadi et al 2021), it can also provide benefits of physical and mental health for residents . Therefore, in the design and renovation practice of urban coastal streets, reasonably increasing the density of trees and vertical greening ensures a high street green coverage ratio. In streets with wide pavements and low interface continuity, boulevards can be created by planting tall trees that can enhance the walk and recreational quality of the streets. Some studies based on perception have shown that the plant community is an important factor influencing aesthetic perception (Iwona 2019), the abundant plant species can effectively enhance the visual quality of streets (Sun et al 2018). In addition, the monotonous street space is brought to life by enriching the hierarchy of plant landscapes (Held 2020). These measures can enhance the attractiveness of the street and improve the human VWIP.
In addition, the water as a unique streetscape element in the coastal street also has a positive contribution to VIWP, increasing the proportion of water can promote the willingness to walk in the coastal streets. Related research has shown that to exposure blue spaces not only benefits physical and mental health (White et al 2010, Gascon et al 2017 but also promote physical activity and social interaction for the residents (Bell et al 2017). For the coastal streets near the shoreline, urban management and designers should play a positive role in the water landscape to promote visual walkability perception for pedestrians. By increasing the interval and reducing the density of trees on the coastal interface. In addition, the negative impact of high-rise buildings on Pedestrian's view can be reduced by controlling the building height on the coastal interface. These measures can create a high quality coastal visual landscape.

The influence of street enclosure on VIWP
The regression results also indicated that buildings have a negative influence on the VIWP. The high proportions of buildings and the continuous building interface create an enclosure street for people's perception. These factors will reduce people's willingness to walk in the streets. Relate studies have shown that denser and taller buildings can create a depressing atmosphere (Zarghami et al 2019, Meng et al 2020, Wang et al 2020, which can impact the pedestrian experience in the streets. In this study, the sky did not show a positive contribution to the VIWP, which is not consistent with previous studies, the result caused by the different background of city. Long et al 's study was based on high-density urban streets of Shanghai, China, in this case, open skies and light may be more attractive to walking behavior for pedestrians. However, in the Fukuoka coastal streets, which have low or middle density of buildings, the influence of the sky on VIWP becomes less obvious. The negative impact of the proportion of buildings on VIWP can be reduced by the construction of small vertical greening. In addition, by reducing the canopy width and increasing the plant distance to create an open green landscape. In commercial streets, it is necessary to control the building interface on both sides of the street to avoid the building interface completely blocking the pedestrian's view.

The influence of security perception on VIWP
The study results showed that cars have a negative influence on the VIWP. Motor vehicles and non-motorized vehicles as dynamic elements in the street space, these elements usually arouse insecurity sense for pedestrians (Xu et al 2018). In urban spaces, people and vehicles are moving together most time, in this case, the crowded and chaotic traffic environment can lead to tension and unpleasant emotions for people. Related studies have proved that wide pavements, as the primary pedestrian infrastructure, encourage walking behaviors by providing a safe and comfortable walking environment (Frackelton et al 2013). Therefore, the planning of urban streets should be as clear as possible, separating sidewalks, non-motorized lanes, and motorized lanes on urban roads to create an orderly street environment. In addition, the overall image of the urban pavement is enhanced by using railings, pavement color and textures depending on the situation. Pedestrian detours can be effectively reduced by optimizing street access, reducing the number of road junctions, and adjusting road fences to improve traffic conditions.

Conclusion
In this study, we used an efficient and accurate method to measure people's VIWP. The proportion of streetscape elements extracted from GSVP images by semantic segmentation, the correlation and regression analysis were carried out between the VIWP evaluation values and streetscape elements. The regression model of the VIWP and the streetscape elements was established.
The results showed that the VIWP of Fukuoka coastal streets has differences in the spatial distribution, the VIWP is better in the western coastal streets, and the eastern coastal street have a poor VIWP. In addition, the VIWP of coastal streets were mainly influenced by three aspects: natural features, street enclosure and safety perception. Natural features including trees, shrubs, grasses, and water have a positive effect on VIWP. Buildings from street enclosures and vehicles from safety perception have a negative impact on VIWP.
This study reveals the influence mechanism of the coastal street environments on VIWP and objectively reflects the spatial quality of the coastal streets, provides accurate data to renewal and design of coastal streets. The results support an urban design theory and practice. It is important to enhance the planning and management of coastal streets, especially to improve the VIWP of the coastal street space in Fukuoka. In addition, the methods were used in this study to efficiently and accurately measure people's visual walkability perception and streetscape elements in coastal streets. With the further development of mapping services, the street view images can be more easily obtained in more area, it can help researchers to understand the impact of the coastal street environment on VIWP in different areas. We believe this study has important implications for building and improving the quality of urban streets.
Although this study explored the influence of coastal street environments on human VIWP, there are still some limitations also worth discussing in future studies. First, the GSVP images are limited by the time when the street view photos were taken. In future study, the samples of the latest updated street view data will be collected for further analysis. Also using historical data from Google Street View images to explore the historical evolution of the coastal street environments, it can help relevant researchers to better understand the influence of urban street environments on VIWP in multiple time dimensions. Secondly, although GSVP images can effectively and accurately predict static street elements, they cannot fully represent the movement elements such as motor vehicles and pedestrians. The study needs to integrate the on-site observation and traffic data to deepen the understanding of VIWP in the future. Thirdly, related research suggests that socioeconomic factors have a significant impact on neighborhood greening, implying that disadvantaged neighborhoods may have lower visual perceptions of walkability, further research on socioeconomic indicators should be added to future studies to explore the influence of socioeconomic factors on human visual walking perception. Finally, the GSVP image data in this study are from the coastal streets in Fukuoka, Japan. Considering the street style and characteristics, vegetation types, urban environment, the regression model may not be applicable to all coastal streets completely.

Data availability statement
All data that support the findings of this study are included within the article (and any supplementary files).