Trend-Residual Dual Modeling for Detection of Outliers in Low-Cost GPS Trajectories

Abstract

Low-cost GPS (receiver) has become a ubiquitous and integral part of our daily life. Despite noticeable advantages such as being cheap, small, light, and easy to use, its limited positioning accuracy devalues and hampers its wide applications for reliable mapping and analysis. Two conventional techniques to remove outliers in a GPS trajectory are thresholding and Kalman-based methods, which are difficult in selecting appropriate thresholds and modeling the trajectories. Moreover, they are insensitive to medium and small outliers, especially for low-sample-rate trajectories. This paper proposes a model-based GPS trajectory cleaner. Rather than examining speed and acceleration or assuming a pre-determined trajectory model, we first use cubic smooth spline to adaptively model the trend of the trajectory. The residuals, i.e., the differences between the trend and GPS measurements, are then further modeled by time series method. Outliers are detected by scoring the residuals at every GPS trajectory point. Comparing to the conventional procedures, the trend-residual dual modeling approach has the following features: (a) it is able to model trajectories and detect outliers adaptively; (b) only one critical value for outlier scores needs to be set; (c) it is able to robustly detect unapparent outliers; and (d) it is effective in cleaning outliers for GPS trajectories with low sample rates. Tests are carried out on three real-world GPS trajectories datasets. The evaluation demonstrates an average of 9.27 times better performance in outlier detection for GPS trajectories than thresholding and Kalman-based techniques.

Joint Stochastic Spline and Autoregressive Identification Aiming Order Reduction Based on Noisy Sensor Data

“…The problem of smoothing data appeared over the years in many contributions within SI field, but they often exhibit limitations or are too specific to some applications [ 20 ]. An early work is [ 21 ], where a smooth cubic spline model is derived from minimizing the sum of squares of deviations for each data segment with first derivative continuity constraints, but the segment lengths are fixed (non adaptively set).…”

“…Still, the data selection cannot be considered adaptive. In [ 20 ] a model-based approach for optimally detecting outliers in GPS trajectories is proposed. To this aim, a cubic smooth spline is firstly estimated to adaptively extract the trend of the GPS observed trajectory.…”

Evaluación de receptores GPS de bajo costo de alta sensibilidad para trabajos geodésicos. Caso de estudio: línea base geodésica

“…Actualmente los equipos de bajo costo también son llamados de alta sensibilidad (Schwieger, 2009;Tsakiri et al, 2016), ya que tienen en promedio una sensibilidad de -160 dB, en contraste a los receptores geodésicos que alcanzan los -128.5 dB (Van Digglen, 2009) ocasionando que reciban señales más débiles (degradadas por edificios o que están por salir del área del receptor). Algunas otras cualidades de estos equipos son el bajo consumo de energía, reciben datos en una frecuencia, proporcionan datos de código, poseen fase portadora y su reducido tamaño los hacen altamente portables (Benedetti et al, 2017;Chen et al, 2016).…”

Forecasting volatility with outliers in Realized GARCH models

“…Until now, many scholars introduced the method of AO and IO detection and correction into the GARCH model and achieved favorable results (Amélie, 2008;Charles & Darné, 2005;Franses & Ghijsels, 1999). Besides, the method is also adopted in other areas, such as GPS tracking data correction (Chen, Cui, & Fu, 2016). However, existing research on Realized GARCH models neglects the AOs and IOs on return series and realized volatility, which may lead to biased parameter estimates and volatility forecasts.…”