5. Travel Time Prediction for Arterial Corridors Using GPS Technologies
Date of Start

August 2005
Indian Institute of Technology Madras (R, C)


Scope and Objectives

The scope of this study was restricted to the study location: part of I. T. corridor, (OMR) Chennai. The location was chosen to represent heterogeneous or mixed flow conditions, in the absence of lane-discipline ,prevalent in urban areas in India. The data was collected for the morning peak hours (2 hours) .The data collection was carried out for a period of one month using the Global Positioning systems (GPS) devices for capturing real time traffic data.

(i) To develop a modeling framework and apply this framework for speed estimation and travel time estimation that integrates real-time and historic traffic data under heterogeneous traffic conditions using Kalman filtering technique

(ii) Identify the influence of key factors affecting performance of Kalman Filter model under heterogeneous conditions

In this study, the prediction was performed using the Kalman Filtering Technique. The Kalman filtering process is a recursive solution technique. Recursive estimation is computationally very quick and thus suited to real-time applications. The Kalman technique was applied with the state variable of interest as the link travel time. Global positioning devices (GPS) offer the promise for collecting traffic data in an inexpensive and non-intrusive fashion, and requires less manpower. Hence, GPS devices were selected for data collection in the study. Bus trips were used as probe vehicles on the study Corridor.

The study corridor was of 14.5 Km stretch (I. T. corridor, Chennai.) was divided into average link length of 0.5 km each. The originating point of the stretch was from Madhya kailash junction and it was destined upto Kumaran Nagar. A total number of 35 trips were made. The Kalman filter technique was applied for prediction with an objective that the kth link travel time can be predicted with prior knowledge of the travel times of the (k-1)th link . For each link the total link travel time i.e., (the time taken for the bus to complete the link) and the stopped time (the time for which the bus was totally stopped and the time for which the bus moved at a speed less than 4.5 kmph (3 miles per hr.) or less was considered as a ‘stop’. This speed was taken from the literature and was also considered to be suited based on the physical difficulties observed in the field in observing the speedometer when the bus traveled at lower speeds. The travel time data for each link for each of the trip was obtained from field. For each trip considered on a given day, data on the total link travel times, the link running time and the stopped times for all the links were calculated. The relative allocation of apriroi and measurement data sets were made with different combinations like continuous and discrete apriori sets.

The apriori and the measurement models were built and from the parameters obtained from these were given as input for the Kalman model. Two separate Kalman models were developed for the link travel time as well as the link running time. Models with the logarithmic of link travel time as state variable were also developed. The performance indices such as Root Mean Squared Error (RMSE) and the Mean Absolute Relative Errors (MARE) were computed to know the performance of the models.
Findings and Conclusions
(The travel time estimates show an RMS error ranging from 9.5% to 12.5%.The performance of Kalman filter for travel time estimation could be improved by using the continuous model, use of travel time instead of logarithmic of travel time as the state variable, and to have separate Kalman Filters for stop time and running times. The study suggests need for and use of significant sample size for the measurement model, since the measurement data appeared to be noisier and played a large role under heterogeneous traffic conditions to improve the accuracy of the Kalman Filter.