利用行人轨迹挖掘城市区域功能属性

Mining Urban Functional Areas Using Pedestrians' Movement Trajectories

城市土地利用功能区是城市规划中的一个重要概念,遥感技术手段在城市土地利用类型识别和动态监测中取得了很大进展。然而,由于城市实际功能的复杂,往往很难从遥感影像中获得城市各个区域的社会、经济或文化等功能属性。互联网技术的发展和移动定位设备的普及,极大地便利了行人移动轨迹数据的获取。本文从行人移动规律和模式与城市功能分区之间高度相关的角度出发,通过机器学习的方法,从大量行人轨迹数据中挖掘隐含的城市功能属性与强度。该方法首先利用矢量栅格化和数学形态学方法,将城市不同等级的路网分割为互不相同的空间单元;其次,根据行人轨迹数据的时空分布特点,定义9个变量并构建高斯混合模型(Gaussian mixture model,GMM),对上述空间单元进行非监督分类,得到7种城市用地类型;随后,结合选定的60个样本区以及人为标识的6种功能区(教育用地、绿地休闲区、一般商业区、政府设施、中心商业区、住宅区),依据样本功能区GPS轨迹时间分布特征,最终对7种城市用地类型进行功能配对;最后,利用核密度估计方法进行功能区强度的可视化。该框架结合机器学习的优势,结果具有较高的准确度。

Urban functional area is an important concept in urban planning. It is difficult to obtain regional functional properties by remote sensing techniques since land uses are not always directly observable even by the closest inspection. With the development of Internet and popularity of mobile positioning devices, a large volume of pedestrians＇ movement trajectories are available to the common. This paper applies machine learning method to discover urban functional areas implicit in the GPS data. The analytic framework is built on an idea that people＇s movement patterns and trip rules have a strong correlation with regions＇ functions. Gaussian Mixture Model （GMM） is applied to extract different functional clusters through iterative machine learning process. About 60 areas are therefore sampled to determine the specific functional areas for residential, commercial, governmental, educational and leisure area. The 6 functional areas are identified with the help of samples＇ temporal distributions. The preliminary results indicate that urban functional areas can be discriminated by integrating GPS movement trajectories with machine learning method, especially with large amount of data. This spatial data mining process is simple, applicable and easy to be carried out in the actual production.