| Big traffic data is an essential ingredient to smart urban transportation,among which the GPS trajectory data is a typical representative.A huge volume of trajectory data of moving vehicles is being accumulated and collected at data centers with unprecedented speed and scale,as a result of the proliferation of GPS-enabled devices and the maturity of Mobile Internet during recent years.However,such massive trajectory data also causes a series of issues to build and realize the smart cities,including heavy communication overhead,storing,and computing and so on.It is well-recognized that online trajectory compression is one of the promising methods to alleviate the above-mentioned issues.At the current stage,existing systems usually achieve data compression by lowering the sampling rate of moving vehicles.Such a naive solution no doubt aggravates the issues of “sparseness” and “uncertainty” in the collected trajectory.To make matter worse,the dense road network and the positioning error of GPS devices will make the inference of true driving routes more challenging.In this thesis,we take a more sophisticated approach to tackle with,i.e.,we develop a novel trajectory mapping and compression system.More specifically,we propose a map matcher,namely SD-Matching(Spatial-Directional Matching),which is proved to 1)have the potential to reduce the side effects caused by the location error significantly;2)generate a more natural and semantical trajectory representation.Next,to save the communication and storing overhead,the trajectory in the spatial and temporal dimensions is compressed based on HCC(Heading-Changes-Compression)and DAVT(Distance-Acceleration-Velocity-Timestamp)algorithms,respectively.Such computation in the system is usually resource-hungry and GPS devices themselves just cannot afford these burdensome tasks(i.e.,trajectory mapping and compression).In addition,there are some requirements that the system implemented in the mobile environment should satisfy,i.e.,computing capability,low latency,and light weight.Motivated by mobile edge computing,the burdensome computation is offloaded to nearby smart devices,such as drivers’ mobile phones.The main contributions are summarized as follow.(1)In trajectory mapping,we first explore the usability of vehicles’ heading direction,and then fully leverage this information in a smart way to improve the efficiency and accuracy of SD-Matching algorithm.(2)In spatial trajectory compression,we first observe the heading changes at intersections.Based on the obtained observation,we then develop a high-quality compression algorithm,i.e.,HCC algorithm,which can achieve a good balance between efficiency and compression ratio.(3)In temporal trajectory compression,we propose a new trajectory representation consisting of three parts,i.e.,distance sequence(D),acceleration and instantaneous velocity sequence(AV)and timestamp sequence(T),and three compressors are devised wisely to compress each part separately.To the best of our knowledge,we are among pioneers to compress the time-varying velocity information,which is essential for driving style classification and urban traffic condition detection.(4)In system implementation,inspired by the principles of mobile edge computing,burdensome tasks are offloaded from GPS devices with limited computing capability to nearby drivers’ mobile phones.Compared to the traditional method which simply collects vehicles’ locations less frequently at the side of GPS devices without compression,it is expected the proposed system is more informative and robust to GPS noises.(5)The extensive evaluation of trajectory matching,trajectory compression,and the system are carried out with real-world datasets.According to the experimental results,we can make a safe conclusion that they outperform similar algorithms and systems in terms of accuracy,compression ratio,and efficiency. |
PDF Full Text Request