Study of travel behavior using Wi-Fi probes

In this study, Wi-Fi probes are used to make data collection easier and more flexible. Wi-Fi probes were used in collecting data from monitoring vehicular traffic and pedestrians. In vehicular traffic monitoring, it was found that maximum detection rate was 34.63%. But some of these MAC addresses detected were from pedestrians as well. In other locations with less pedestrians, the detection rates were very low. The results from origin-destination (OD) analysis show that the number of trips between zones were very small as compared to the number of detections of vehicles in different locations. And the average detection rate of the overall detection from different zones was 14.24%. This also indicates that Wi-Fi probe can detect pedestrians better than vehicle. In the second part, human population density travel behavior, it was found that number of MAC addresses detected are very high when compared to field surveys data due to MAC address randomization. Thus, an experiment on human population density was conducted to find a relationship between number of detections (NoD) and number of people (NoP) in the area, to use this relation to predict number of people, and to find out other factors which affect number of detections. The relationship between NoD and NoP was represented using linear regression model. The average error from model calibration was 20%. It is also found that as NoD increase the error from model is also lower. Other factors affecting data collection include mobile modes of operation, which lower NoD, and Wi-Fi connection status, which may increase noises in NoD. By applying this information in removing noisy data, it’s possible to analyze for travel behavior information of people in the study area. This include tendency of trips from different locations and trip patterns of the people. In conclusion, Wi-Fi probes can detect pedestrian better than vehicles. Due to limitation of the device, Wi-Fi probes are suitable for collecting a sampling of information, about travel behavior, from the whole population. Wi-Fi probes are also suitable for collected matched MAC addresses for OD estimation in long term data collection.

本研究采用Wi-Fi探针（Wi-Fi probes）以实现更便捷灵活的数据采集，利用其完成车辆与行人交通监测的数据采集工作。在车辆交通监测场景中，最大检测率可达34.63%，但部分检测到的媒体访问控制（Media Access Control, MAC）地址同时属于行人；在行人较少的其他监测点位，检测率则极低。起讫点（Origin-Destination, OD）分析结果显示，相较于不同点位的车辆检测量，区域间的出行总量极低；各区域整体检测的平均率为14.24%，这也表明Wi-Fi探针对行人的检测效果优于车辆。在第二部分的人口密度与出行行为研究中，由于媒体访问控制地址随机化（MAC address randomization）技术的应用，检测到的MAC地址数量远高于实地调研数据。为此，本研究开展了人口密度相关实验，旨在探究区域内检测数（Number of Detections, NoD）与实际人数（Number of People, NoP）之间的关联，利用该关联实现人数预测，并挖掘影响检测数的其他因素。研究采用线性回归模型表征NoD与NoP之间的关联，模型校准的平均误差为20%，且检测数越高，模型预测误差越低。影响数据采集的其他因素包括移动工作模式（会降低检测数）以及Wi-Fi连接状态（可能增加检测数中的噪声）。通过利用上述信息剔除噪声数据，即可对研究区域内人群的出行行为特征展开分析，包括不同点位的出行倾向与人群出行模式。综上，Wi-Fi探针对行人的检测效果优于车辆。受设备本身局限性影响，Wi-Fi探针仅适用于从全人群中采样获取出行行为相关信息，同时也适用于长期数据采集场景下的匹配MAC地址以开展起讫点估计工作。