Efficient Algorithms for Vehicle Type Identification Using Mobile-Phone Locations

Accurately estimating traffic flow parameters is a key technology for intelligent transportation systems (ITS) which services to increase the safety, efficiency and reliability of the transportation system. Recently, monitoring the traffic flow using mobile phones data from wireless telecom operators has shown to be promising. However, the difficulty lies in identifying the vehicle type of the mobile phone holder. In this paper, we propose an agglomeration clustering algorithmto classify all the phones detected on a observed freeway into a number of clusters such that each cluster indicates a travelling vehicle. In the algorithm, each phone is initially treated as a cluster, then highly similar clusters are merged into one as they come from the same vehicle. After clustering analysis, the vehicle type of each cluster is recognized based the counts and speed of phones within the cluster. Different from previous work, in case of phone location data missing, we use the incomplete original data rather than estimating the missing information by interpolation methods, which avoids the errors caused by data interpolation. Experimental results show that our approach significantly improves the previous work, and the improvement is more obvious on datasets with missing data.