Multisource And Heterogeneous Traffic Data Fusion Research Based On Gaussian Mixture Model

Because of the increase in vehicle usage and insufficient of road capacity, traffic congestion becomes more and more serious. The fundamental principle to solve the problem is to decrease the density of the vehicles on the road. One method to achieve this goal is to estimate the real time traffic state and guide the drivers aiming at increase the road utilization. In order to obtain more accurate traffic state, various traffic data is generally fused and a fusion model based on Gaussian Mixture Model (GMM) is proposed. This model can overcome the ambiguity of the single-source sensor efficiently and estimate the traffic state of the urban road network accurately. Based on the existing research results, this paper is focused on preprocessing of sensor data, optimizing of the road speed from video source and data fusion of the traffic data, and the research contents and original contributions are as follows:Firstly, for the external factors will bring down the quality of traffic data, this paper takes the road level into consideration and propose a method utilizing corresponding thresholds and kalman filter to filter the traffic data. Secondly, because the traditional traffic sensors area sensitive to the change of outside illumination, this paper utilizes RBG-D cameras to get the RGB information and the Depth data which means the distance to the target at the same time to obtain more scenario information in order to increase the accuracy of vehicle identification and tracking and finally increase the accuracy of road average speed estimation. At last, for the lack of reliability of single-source sensors will decrease the accuracy of traffic state estimation, this paper takes multiple traffic sensors into consideration, utilizing the GMM to model heterogeneous data and proposes a fusion method based on GMM to improve the accuracy and robustness of traffic state estimation.With the RTMS, GPS and RGB-D data collected from four road segments in Hangzhou from June 7 to 12,2015, this paper takes a demonstration study on the proposed fusion model based on GMM and arrives at the following conclusions:(1). The robust correction of the traffic data which combines the road level is of positive significance in data quality. (2). Depth data can make up for the loss of the color information when the illumination is weak and the road average speed estimation is more accurate. (3). The proposed fusion method based on GMM which takes the distribution of traffic flow into account can increase the robustness of traffic state estimation efficiently.