基于深度学习的商业综合体空间超链接分析——以三里屯太古里和北京APM为例  

Research on Spatial Hyper-Links in Commercial Complexes Based on Deep Learning:A Case Study of Tai Koo Li Sanlitun and Beijing APM

作　　者：吴韶集 胡一可[1] WU Shaoji;HU Yike

机构地区：[1]天津大学建筑学院

出　　处：《南方建筑》2022年第1期61-68,共8页South Architecture

基　　金：国家自然科学基金重点项目(52038007):基于中华语境“建筑-人-环境”融贯机制的当代营建体系重构研究。

摘　　要：实测人流量数据在当前的建筑学研究中被广泛采用,其可用于建立具有预测能力的空间模型。使用基于深度学习的Yolo-v3和DeepSORT算法获取实测人流量数据;而后基于空间句法理论分析了电梯和自动扶梯两种不依赖人自然运动路径对于三里屯太古里和北京APM中人群分布的影响。结果显示这两种路径对人群分布产生影响,产生了"空间超链接"现象。这一结果为建筑师进行建筑实践提供了一定依据,合理选择和布置自动扶梯和电梯可影响建筑中的人群分布产生情况。In previous studies concerning architecture,spatial models with predictive capabilities were developed using the measured pedestrian flow data.The spatial hyperlink is a concept that explains the connectivity mechanisms of modern urban transport systems.Specifically,people can travel directly from one space to another through the increasing non-human natural movement paths.At the architectural scale,paths with non-human natural movement modes,such as lifts and escalators,also agree with the definition of spatial hyperlink paths.Their impacts on the spatial systems within buildings are not known yet.This study aims to address two questions.Firstly,quantifying the effect of spatial hyperlinks on the distribution of people in a commercial complex,and secondly,what are the differences between different types of spatial hyperlinks.This study is undertaken in three steps.The commercial complex’s path system is transformed into a network model,and its betweenness centrality is calculated.Videos at selected observation points on the path of the study area are then recorded,and a machine learning-based algorithm is applied to the recorded data to count pedestrian flows.Deep learning is a particular type of machine learning algorithm that enables non-linear information processing by building a multi-layer perceptron structure using a computer.A multiple linear regression analysis was applied on the measured pedestrian flows,and the spatial hyperlinking phenomenon was examined and interpreted.The programming utilised Python,using Tensorflow and Keras as the deep learning package,OpenCV as the computer vision package,Pandas as the data processing package,and Statasmodel as the statistic package.The spatial hyperlink phenomenon was investigated through multiple linear regression analysis with Taikoo Li Sanlitun and Beijing APM.Considering the two research questions,the following conclusions were apparent:(1)Both the lifts and escalators in the two research cases influence pedestrian flows;(2)the phenomenon of spatial hyperl

关 键 词：深度学习 计算机视觉 空间句法 商业综合体 超链接 

分 类 号：TU247[建筑科学—建筑设计及理论]