17:15 〜 18:30
[HCG25-P04] Urban Park Visitor-Mapping using Google Street View Panorama
(Case Study: Shinjuku-Gyoen during Spring Season)
キーワード：behavior mapping, big data, panorama image, people count, sky view factor
Hanami (literally means "flower viewing") is the Japanese tradition of enjoying the beauty of flowers; but, it mostly refers to cherry blossom ("sakura" in Japanese) viewing in spring. People usually sit on a tarp beneath sakura trees and having a picnic. People go to popular spots for hanami during the blooming period, such as rivers and parks. Shinjuku Gyoen is a large park and garden in Shinjuku and Shibuya, Tokyo. It has more than one thousand sakura trees blossom during the spring season, making it one of Tokyo's most popular and pleasant hanami spots.
Behavior mapping is a useful tool for observing the interaction between people and places. The process usually requires a lot of resources from cost, time, and staffing. More studies are trying to do behavior mapping more efficiently. Google Street View (GSV) is a component of Google Map and Google Earth that provides street view images captured in many cities worldwide. This study aimed to map visitors' distribution in Shinjuku Gyoen during the hanami season efficiently using GSV image and analyze the factors that determine the visitor's distribution in the park.
The data analyzed in this research are geotagged panoramic images collected from GSV using software named "Street View Download 360" developed by Thomas Orlita. A total of 1058 panorama images in Shinjuku Gyoen during spring were analyzed. Then, we did the people count analysis to estimate the number of visitors using an automatic image recognition OCR (Optical Character Recognition) API (Application Programming Interface). The application is available in Aliyun Market, provided by Sichuan Fuqing Big Data Technology Co., Ltd. The results of this people count analysis were then mapped to see the visitor's distribution. To analyze the factors that determine the visitor's distribution, we tested four factors: sky view factor (SKF), distance to the park gate, distance to water bodies, and distance to rest house facilities. We used Spearman correlation analysis and linear regression analysis methods to describe the relationship between the number of people and the analyzed factors.
The results show that a total of 6,206 people was counted in the park. The range of people count in each point varies from 0 to 49 people. We found that SVF value and distance to rest house have high correlation with people distribution from the correlation analysis. SVF has a positive correlation (0.37), indicating that more people tend to be found at spots with higher SVF or high sky visibility. Meanwhile, distance to rest house shows a negative correlation (-0.44), which means the closer a spot's distance to the rest house, the higher the number of people. Distance to park gate and water bodies show a low correlation of 0.04 and -0.01, respectively.
We concluded that park visitors are most likely to be around the open space area (places with high sky visibility) and near the rest houses during the spring season. Then, the results of people count probably underestimate the actual number because, in really crowded areas, the API might fail to detect people in the far. However, we think this factor does not change people's tendency to gather in open spaces during the spring season. This study developed a generalizable approach using GSV for people mapping in the urban parks and demonstrated its applicability using a case study of Shinjuku Gyoen Park. The methods in this research are useful to remotely analyze people's behavior in a big-sized area as it is more time- and cost-efficient. Future research is encouraged to do a similar analysis in the other parks in Tokyo in the same season to see if the behavior is also the same.
Behavior mapping is a useful tool for observing the interaction between people and places. The process usually requires a lot of resources from cost, time, and staffing. More studies are trying to do behavior mapping more efficiently. Google Street View (GSV) is a component of Google Map and Google Earth that provides street view images captured in many cities worldwide. This study aimed to map visitors' distribution in Shinjuku Gyoen during the hanami season efficiently using GSV image and analyze the factors that determine the visitor's distribution in the park.
The data analyzed in this research are geotagged panoramic images collected from GSV using software named "Street View Download 360" developed by Thomas Orlita. A total of 1058 panorama images in Shinjuku Gyoen during spring were analyzed. Then, we did the people count analysis to estimate the number of visitors using an automatic image recognition OCR (Optical Character Recognition) API (Application Programming Interface). The application is available in Aliyun Market, provided by Sichuan Fuqing Big Data Technology Co., Ltd. The results of this people count analysis were then mapped to see the visitor's distribution. To analyze the factors that determine the visitor's distribution, we tested four factors: sky view factor (SKF), distance to the park gate, distance to water bodies, and distance to rest house facilities. We used Spearman correlation analysis and linear regression analysis methods to describe the relationship between the number of people and the analyzed factors.
The results show that a total of 6,206 people was counted in the park. The range of people count in each point varies from 0 to 49 people. We found that SVF value and distance to rest house have high correlation with people distribution from the correlation analysis. SVF has a positive correlation (0.37), indicating that more people tend to be found at spots with higher SVF or high sky visibility. Meanwhile, distance to rest house shows a negative correlation (-0.44), which means the closer a spot's distance to the rest house, the higher the number of people. Distance to park gate and water bodies show a low correlation of 0.04 and -0.01, respectively.
We concluded that park visitors are most likely to be around the open space area (places with high sky visibility) and near the rest houses during the spring season. Then, the results of people count probably underestimate the actual number because, in really crowded areas, the API might fail to detect people in the far. However, we think this factor does not change people's tendency to gather in open spaces during the spring season. This study developed a generalizable approach using GSV for people mapping in the urban parks and demonstrated its applicability using a case study of Shinjuku Gyoen Park. The methods in this research are useful to remotely analyze people's behavior in a big-sized area as it is more time- and cost-efficient. Future research is encouraged to do a similar analysis in the other parks in Tokyo in the same season to see if the behavior is also the same.