Comparing conventional manual measurement of the green view index with modern automatic methods using google street view and semantic segmentation

Urban greenery has various beneficial effects, such as engendering peace of mind. The green view index (GVI) effectively measures the amount of greenery people can perceive and is a suitable indicator of urban greening. To date, the most common way to measure the GVI has been to photograph the street environment from eye level and use image-editing software to calculate the area occupied by vegetation. However, conventional methods are time-consuming and labor-intensive, and the calculation results may vary among individuals. In recent years, the use of Google Street View (GSV) photos and calculation of the GVI using automatic image segmentation have rapidly developed. In this study, we demonstrate the advantages of GSV and image segmentation over conventional methods, verify their accuracy, and identify the shortcomings of modern methods. We calculated the GVI in the central part of Sapporo, Japan, using the automatic image segmentation AI “DeepLab” and compared the results with those measured by Photoshop. At the exact GSV locations, we also acquired photos and again calculated the GVI using AI, subsequently comparing the results with those obtained on-site manually. Although the correlations were high, automatic image segmentation tended not to identify lawns and flowers planted in the ground as vegetation. It was impossible to determine the year when the GSV photos were taken. In addition, the distance to greenery was biased, depending on the position on the street. These points should be considered when using these modern methods.