Urban Traffic Monitoring Based On VANET And Smartphones

Trafc congestion, which leads to the wasting of time and fuel, has become asevere problem in city modernization. Trafc information, i.e. trafc condition of eachroad, has shown to be efective and necessary to reduce trafc congestions in cities.This is because according to the trafc information, trafc monitoring center (TMC)can control the trafc lights and provide trafc guidance while drivers can plan theirroutes to avoid congesting roads. Due to the importance of trafc information, thisthesis focuses on monitor trafc conditions in urban area efectively based on mobilesensor network (MSN).Generally, the trafc conditions of the roads are collected by the static sensors,e.g. loop detectors underground and video cameras. However, this traditional methodsufers from high cost on deployment and replacement. Consequently, beneft from themobility of the sensor nodes, it has been proposed to utilize MSN to sense and collec-t the trafc conditions. In the MSN-based trafc monitoring system, probe vehicles(PVs) running in the city and smartphones act as mobile sensor nodes, and upload thesensing reports to the TMC, where trafc information is estimated and published.The main contributions are summarized as follows.1. According to the analysis on the real traces of about30,000PVs in Shanghai,we discover that the sampling rate of trafc conditions is only about25%andthe un-sampled trafc conditions require to be estimated. As the matrix of thetrafc condition can be approximated to a low-rank matrix, matrix completion(MC) based method is used for estimation. To accelerate the computation, aL1/2-norm based method is introduced in the algorithm. Moreover, as we haveobserved that the trafc conditions of the roads usually doesn’t change sharply except for the rushing hours, a cost function is added to the objective functionto punish the large changes among two consecutive samples at the same road.Finally, due to the fact that the elements in the trafc matrix are non-negative, weintroduce a cost function to punish the violation of the bound. Based on the ideaabove, an algorithm named HaTTEM is proposed to estimate the un-sampledtrafc condition in the trafc matrix.2. To further improve the accuracy of the estimation, we also fnd that the perfor-mance of matrix completion is infuenced by the sampling process, based on theanalysis of real traces. Due to the uneven distribution of the PVs, about20%of the roads have no reports for a whole day, which leads to large estimation er-ror. We propose to control the moving of the foating cars from TMC to enhancethe performance. The relationship of the estimation error and the sampling pro-cess is modeled from the prospective of entropy. Based on the relationship, thesampling rule for the foating cars are achieved. Further, we develop the patrolalgorithms for the foating cars.3. To achieve more trafc reports, we propose that we can also use smartphones assensor nodes to collect and upload sensing reports when the users are inside thevehicle, as the moving speed can be achieved from the GPS in the mobile phone.To achieve this, the smartphones should be able to automatically determine theuser is inside the vehicle or walking. As the vertical acceleration of walkingis diferent from that of sitting or standing inside the vehicle, we calculate thevertical acceleration from the reading of accelerometers and gyrometers in thesmartphones, and the operation of coordinate transformation. A simple AndroidAPP is developed to collect experiment data, based on which a simple classiferis proposed and the parameter is determined.In the end, the thesis is concluded and the future works are discussed.