414 / 2019-02-27 12:19:18

A Fast Trajectory Filtering Method for Massive GPS Data Based on Neighboring Points

GPS Stream,GPS Trajectory,Outliers,Noise Filtering,Algorithm

全体主题 > Spatial data collection, Analysis, and Visualization

With the widespread application of wireless communication technologies in logistics management system, massive GPS trajectory data are generated. The precise spatio-temporal position data of vehicles can be of great value for urban planning and logistics operation. It is also the research hotspot of “Smart City”. However, on account of sensor noise, inner noise of receivers and some other factors, data provided by the existing Global Positioning System (GPS) inevitably incorporates occasional outliers. To smooth the outliers in massive data efficiently and accurately for subsequent use, an improved median filtering method based on neighboring points is proposed.

The method is composed of two stages: preprocessing and filtering. In the preprocessing stage, duplicate records are abandoned. Also, the consecutive static points in the trajectory sequence are replaced by a single point. In the filtering stage, non-boundary points and boundary points are separately processed. Consider the window size as 2k+1. Each 2k+1 consecutive points can be classified into two categories: 1) straight lines; 2) broken lines which consist of sharp turns in the data sequence. As for non-boundary points in straight lines, the predicted coordinate is the median of latitude and longitude of k neighboring points. While those in broken lines, where the maximum change of azimuthal angle between two neighboring points is larger than a given threshold, the parameter k will be persistently divided to its half until the segment becomes a straight line or k equals to 1. Then the procedures of processing straight lines are applied to predict the points in broken lines. As for boundary points, the predicted coordinates and threshold are used to determine whether the boundary points are in the reasonable range. If not, their coordinates will be modified according to the azimuthal angle.

To be more convincing, several experiments are carried out to prove this method. We contrast this method with Kalman filter and traditional median filter on the 2016 GPS data of truck trips that started or ended in Shenzhen provided by related urban authorities. The dataset is composed of more than 10 billion records and six different attributes, including license plate, longitude, latitude, time, azimuthal angle and speed.

The experiments show that the running time of our improved median filter is similar to the traditional median filter, and much shorter than Kalman filter. Compared with the traditional median filter and our improved median filter, Kalman filter suffers a greater sensitivity to outliers in general. In addition, compared with our improved median filter, traditional median filter is less reliable when outliers appear at sharp turns.

In conclusion, we propose an improved median filter which can achieve good performance in filtering trajectories with outliers in large-scale raw GPS trajectory data at low costs. After being processed, the GPS data can be more reliable for further research. With smoothed GPS data, urban planners could mine more accurate activity patterns of humans and vehicles, as well as make better decisions in traffic control and logistics management.