Origin Destination Matrix for Smart City and user behavior analysis

DISIT Origin Destination Matrix is a tool for analysing Smart City flows and user behavior. In the following an example applied on San Francisco data and territory. The innovation consists of producing the OD matrix on the basis of the measures performed on.... ask to Paolo Nesi, the DISIT lab coordinator paolo.nesi@unifi.it
 

The San Francisco Case:

the numbers in the flows refer to the CAPs that can be identified in the following (partial) map.


Thanks to data coming from: Michal Piorkowski and Natasa Sarafijanovoc-Djukic and Matthias Grossglauser, A Parsimonious Model of Mobile Partitioned Networks with Clustering, The First International Conference on COMmunication Systems and NETworkS (COMSNETS), 2009.

         
San Francisco OD matrix as chord among the 13 central ZIP areas of the city (real cab flows)

a)     Random APs: identification of the streets with the highest trace flow rate (those that have at least 30,000 traces) and then randomly selecting 400 APs from the AP grid described above (see Figure 6a for the OD matrix). This set of APs is a subset of the set selected in case (b).  
b) High Traffic APs: identification of the streets with the highest trace flow rate (those that have at least 3000 traces) and then selecting all the APs intersecting those traces, thus resulting in 804 APs (see Figure 6b for the OD matrix).
c)     High Traffic APs (zip boundaries): identification of the streets with the highest trace flow rate (those that have at least 3000 traces) and then starting from the 804 APs of case (b) and considering only those within 300 m of the zip boundaries, thus resulting in 448 APs (see Figure 6c for the OD matrix). This set of APs is a subset of the set selected in case (b).
d)     High Traffic APs (top 400): identification of the streets with the highest trace flow rate (those that have at least 3000 traces) and then starting from the 804 APs of case (b) and considering only the top 400 APs (see Figure 6c for the OD matrix). This set of APs is a subset of the set selected in Case b.
e)     Real augmented with selected High traffic APs: the real distribution of the AP in San Francisco’s downtown was integrated with the top 300 AP from case (d) with the highest traffic rate. This set was then cleaned up by removing those APs that were found to be at a distance less or equal than 50 m from the real APs, and removing also intersecting APs, thus resulting in 400 APs (221 real APs, 179 selected high traffic APs).  

Considering public Wi-Fi access points is not reliable in determining an effective model for the city traffic flows. Instead, placement of sensors only in proximity to the boundaries of high traffic routes yields a statistically significant linear relationship, between the OD matrix trip of sampled and real data. Thus, it is possible to model the traffic behaviour with a limited amount of traffic counters (in this context Wi-Fi access points) with an acceptable precision. The proposed methodology is general and can be applied to different urban scenarios, in the context of Smart City traffic management. It makes use of Wi-Fi access points (AP) distributed across the city of San Francisco. Comparative analysis has shown an increased precision of the AP positioning validated by obtaining an OD matrix and assessing correlation of 0.89 (case (e)) with the actual OD taken from the original real data by considering cab flows from high traffic zones, positioned in proximity of the zones boundaries, with respect to random uniformly distributed or real APs (e.g., restaurants, hotels, schools). The approach allows to identify which are the strictly needed AP to be added with respect to the AP that can be already in place in the city, to exploit the whole infrastructure of Wi-Fi also for people flow monitoring and assessment.