著者
宗本 晋作
出版者
日本建築学会
雑誌
日本建築学会計画系論文集 (ISSN:13404210)
巻号頁・発行日
vol.72, no.618, pp.173-179, 2007
被引用文献数
2 3

The purpose of this paper is to provide the method for the construction of the probabilistic model of preference for space. Bayesian networks are expected to construct probabilistic models including an uncertainty of human behavior for prediction and decision-making. We applied Bayesian networks to construct graphical models that represented the correlation between preference for space and spatial elements of which exhibition was composed. The difference of preference for space was easily understood by visual analysis of graph structures. By executing probabilistic reasoning of Bayesian networks on these models, furthermore we deduced the combination of spatial elements that are expected to be preferred in high probability.
著者
大野 耕太郎 山田 悟史 宗本 晋作
出版者
日本建築学会
雑誌
日本建築学会計画系論文集 (ISSN:13404210)
巻号頁・発行日
vol.87, no.798, pp.1602-1611, 2022-08-01 (Released:2022-08-01)
参考文献数
17

This study aimed to estimate human willingness to visit cityscape images via artificial intelligence (AI) using multimodal deep learning. In this study, gaze information was acquired through subject experiments using a measurement device. We added gaze information when humans felt motivated to visit the cityscape image, and confirmed whether the estimation accuracy of AI would improve. We also created an AI model that generated gaze-view images, and used it for multimodal deep learning. We used pix2pix to generate the images. Finally, we verified the accuracy of the proposed multimodal deep learning approach, when the generated pseudo-gaze image was attached.
著者
山田 悟史 藤井 健史 宗本 晋作
出版者
日本建築学会
雑誌
日本建築学会計画系論文集 (ISSN:13404210)
巻号頁・発行日
vol.81, no.727, pp.2083-2093, 2016 (Released:2016-09-30)
参考文献数
29
被引用文献数
5 7

In recent years, there have been a number of social initiatives related to improving the environment in city landscapes. Green space is becoming a tool to enhance the comfort of city space. "The Basic Plan for Green of Kyoto City" is one such example, where the ratio of visible green space is being used as a tool to improve the city environment. Many studies are being carried out to support this initiative and this study is one of them. The purpose of this study was to: 1) present a method to measure the location/angle specific ratio of green spaces in the omnidirectional visibility rate using a three-dimensional model of the target location, 2) create a perception deduction model based on Self-organizing Maps and 72 variables of visible green space in omnidirectional visibility rate, and 3) statistical verification of the accuracy of the perception deduction model. There are 72 categories of green space in the omnidirectional visibility rate. These categories are based on the location- and angle specific ratio of these spaces. Six of these categories were used for the location specific measurement, namely, "tall trees", "medium trees", "shrubs", " implantable ", "ground cover", and "others". Twelve angle specific measurements for every fifteen degrees were used and eight perception estimation parameters were selected. The perception estimation parameters included: “many or less", "satisfied or not satisfied", "pleasant", "serene", "covered (wrapped)", "close by or far", " surrounded by", "refreshing” and “widely spread". In this paper, we present results from the "ratio of visible green space in the omnidirectional visibility rate map”, the “self-organizing map" and the "perception estimation value map”. During the verification of the perception estimation model (the primary objective of this study), we compared the estimated perception values with the survey based observed values associated with a location of green space that was not included in the model creation. When we compared them statistically, we confirmed a significant correlation (n=32, p<0.05) between the estimated values and observed values (Pearson's correlation). We noted that the strength of the correlation was moderate but significant (correlation coefficient values around 0.6), with when we used the lower significance level (p<0.001). Taking into account effect size from psychological statistics, the average difference between the estimated and observed values of perception can be considered small for the parameters "many or less" "satisfied or not satisfied", "pleasant", "serene", "covered (wrapped)", "close by or far" and" surrounded by". However, the average difference was moderate for “Refreshing” and “Widely spread” and a significant difference between observed and estimated perception values was noted for these parameters in a paired t-test. Consequently, this perception deduction model is able to predict low and high values of “Refreshing” and “Widely spread", however, we need to be aware of the one degree difference, which happens to be the width of the confidence interval and may affect the estimated values.