Inferring Important Locations With Urban Surveillance Camera Data

Nowadays,with the rapid development of various intelligent devices(e.g.,vehiclemounted GPS,smart phones,surveillance cameras)and the widespread popularity of social networks(e.g.,WeChat,Weibo,and Twitter),people's spatiotemporal data can be collected from a variety of data sources.Inferring important locations for users based on spatiotemporal data has become increasingly important for real-world applications ranging from transportation scheduling,urban planning,recommender system,and surveillance camera security.Therefore,we propose some methods to infer important location based on urban surveillance camera data,and the specific work is as follows:(1)We propose an important location area inference method for vehicles based on urban surveillance camera data.First,we obtain a context-rich road network with surveillance cameras.Second,based on the spatial information of the data,we detect the important location areas for each vehicle by clustering Origin-Destination(O-D)extracted in vehicle's camera trajectories.Then,based on the time information of the data,we obtain the 4)9)/ time pattern of each cluster,and according to the different visit time patterns of vehicles in different areas,different important location areas of vehicles are inferred.(2)Based on the above inferring of the important location area,we further propose a work location inference method.First,using the pattern feature of people's working time,the working area is matched from multiple important location areas of the vehicle.Second,by using the obtained road network and the distribution of POIs in the work area,we extract the vehicle's reachable POIs to further reduce the range of work location.Finally,Extensive experiments and case studies on a large-scale real-world traffic dataset in Jinan for a whole month demonstrate the effectiveness of our proposed method.(3)According to the inferring of the important location area,we propose an accurate home location inference method based on sparse urban surveillance camera data.First,based on the time pattern feature of people visiting home location area,we match the home location area from multiple important areas.Second,we propose a Kernel Density Estimation(KDE)based inference method with a local camera selection strategy to effectively identify the home community from the residential communities near/in the home area candidate.Finally,our comprehensive experiments on a large-scale real-world dataset in Jinan for a whole month demonstrate the effectiveness of our proposed method.