Visit the Yahoo! Group for this interest group
Send mail to the local members

The objective of this project is "to establish a methodology for combining (GPS) data from both the Video Log and yearly road condition survey to optimally or near optimally estimate a 3-D spatial model of the Kansas Highway Network which can be adapted for automated cross-referencing to convert between lat/long, the standard KDOT LRS, and other geo-referencing systems used throughout KDOT." We will focus on the following issues that go towards the fulfillment of this objective:

1. Kansas Highway 2D Spatial Modeling Enhancement Based on GPS Lat/long Under Uncertainty Due to the fact that multiple data sets come from different years of the yearly road condition survey and different surveys - the Video Log GPS index data, the current modeling of Kansas Highway adopted a piece-wise polynomial fitting with adjusted weights for each data series to achieve accuracy in defining load segments. The uncertainty in model estimations also came from GPS outliers and unavoidable human errors such as passing other vehicle during survey, riding off road before turning into a new route, etc. While the current modeling approach does a good job in estimating straight road segments, it would over- or under-fit a curve or turn. The issue here is to develop an alternative modeling approach or provide an enhancement to the current approach or both to increase the fitting accuracy of 2D spatial modeling under uncertainties.

2. Feasibility Study of Extending the 2D Model into 3D Model The current spatial model is a 2D representation of Kansas highways based on GPS lat/long coordinates. The issue is whether current data structures can adequately and efficiently support 3D representation. In addition, any modification or enhancement of the current model needs to address this issue as well.

3. Stopping Sight Distance Analysis Based on the Proposed Model We propose to develop an analytical (geometric) model for identifying and visualizing stopping sight distance. This shall be based upon intersection testing using maximum likelihood estimation (MLE) and interpolation methods such as those currently being employed by KDOT. We will develop a prototype visualization system in MATLAB and investigate more effective and efficient graphical user interface and visualization techniques, with the goal of developing a standalone (OpenGL or Java3D) visualization system for GPS-based highway modeling applications.

The following are the tasks that address the issues listed above:

Establish GPS Lat/Long/Elevation Error Range/Distribution

Current GPS technology provides accurate absolute location of a fixed landmark. While longitude and latitude positions are more accurate than the altitude one. In this project, multiple GPS data observations on the same road segment were obtained from survey vans that passed through Kansas highways. Many factors leading to errors of accurate positioning of a road segment include different starting points, different driving speeds, different lane positions of a driving van, possible passing of slow moving vehicles, etc. This task is to access the error distributions of GPS coordinates due to these factors in the content of the proposed project. Note that the error distributions are a combination of all the external factors plus the error caused by GPS measurement. The results from this task will help establish the accuracy of all the subsequent modeling and applications.

Enhance Existing 2D Models and Data Sets

The existing geometric modeling is based on model fitting of a third-order polynomial in a piecewise fashion. While this approach provides reasonable results on most segments of Kansas Highways, it does not work well with some road segments with no-linear shapes, such as 90-degree turns. The cubic polynomial models tend to over- or under-fit the turns. This task is to study other geometric model fitting approaches that may find an application in this case. Tentative areas for this task include geometric presentations in Computer Aided Design related to either manufacturing or architecture. At this stage, we plan to use the current piecewise cubic polynomials as the base for majority of geometric representation. In addition, an intelligent agent shall be applied to detect "trouble spots." Once such spots are detected, the system will switch to a new geometric modeling scheme.

Develop Stopping Sight Distance Analysis and Model

According to AASHTO - Geometric Design of Highways and Streets the "sight distance" is the length of roadway ahead visible to the driver. The minimum sight distance available on the roadway should be sufficiently long to enable a vehicle traveling at or near design speed to stop before reaching a stationary object in its path. Stopping sight distance is the sum of two distances: the distance traversed by the vehicle from the instant the driver sights an object necessitating a stop to the instant the brakes are applied and the distance required to stop the vehicle from the instant brake application begins. Based on the literature review conducted by AASHTO prior to 1990, the reaction time of 2.5 sec is considered to be large enough to include reaction time for nearly all drivers under most highway conditions. KSU research team will review the most recent literature related to the reaction time of the drivers under varies road conditions. The approximate braking distance of the vehicle on a level roadway could be calculated from the following formula:

d = V2/30f where:

d - braking distance, ft

V - initial speed, mph

f - coefficient of friction between tires and roadways

Combine Multiple Data Sets Under Uncertainty/Data Fusion

Multiple observations were obtained for the same road segment. Some data sets were obtained from KDOT's LRS while others are GPS data from pavement management surveys since 1997 and Video logs (three year cycle). These multiple data sets often do not superimpose each other due to different starting points, human errors, equipment errors, speeds of the survey vans, etc, just to name a few. This task is to establish an approach to integrate these multiple data sources to provide cross-references for data accuracy. There are several possible strategies for this task. All feasible solutions will be evaluated and the most suitable one will be chosen.

Develop a Prototype for Extending Model From 2-D to 3-D

The current 2D model is developed in MATLAB using a data structure stored locally on a single PC workstation. This task is to investigate whether and how we can augment the current 2-dimensional model with a third dimension, namely, elevation, and its related attributes. First, we shall determine the scope of the model revision. Second, we shall develop an experimental prototype that provides rudimentary 3-D functionality. This task is strongly correlated with enhancing existing 2D models and Data Sets.

• William H Hsu
• Shing I Chang
• Margaret Rys
• David Ben-Arieh
• Stan Young (Affiliate)

• Sreerama Valluri
• Guoxi Zhang
• Girish Nehate

Back to the KDD Lab main page