Vehicle classification data especially trucks has an important role in both pavement maintenance and highway planning strategy. An advanced microelectromechanical system (MEMS) accelerometer for vehicle classification based on axle count and spacing was designed, tested, and applied to the pavement. Vehicle-pavement interaction was collected by the vibration sensor while vehicle axle count and spacing were calibrated later. Collected vibration data also used to analyze the pavement surface condition and compared with simulation using dynamic loading analysis. Laboratory tests using MMLS3 device to verify the accuracy of MEMS accelerometer and reaction under different surface condition were tested. An algorithm for calculating axle spacing and axle count was developed. Acceleration of different pavement surface condition were analyzed and compared with simulation results, the influence of surface condition to the pavement acceleration was concluded. / Master of Science

No description available.

A real-time Vehicle Make and Model Recognition (VMMR) system is a significant component of security applications in Intelligent Transportation Systems (ITS). A highly accurate and real-time VMMR system significantly reduces the overhead cost of resources otherwise required. In this thesis, we present a VMMR system that provides very high classification rates and is robust to challenges like low illumination, occlusions, partial and non-frontal views. These challenges are encountered in realistic environments and high security areas like parking lots and public spaces (e.g., malls, stadiums, and airports). The VMMR problem is a multi-class classification problem with a peculiar set of issues and challenges like multiplicity, inter- and intra-make ambiguity among various vehicles makes and models, which need to be solved in an efficient and reliable manner to achieve a highly robust VMMR system. To reliably overcome the ambiguity challenges, a global features representation approach based on the Bag-of-Features paradigm is proposed. We extract key features from different make-model classes in an optimized dictionary, through two different dictionary building strategies. We represent different samples from each class with respect to the learned dictionary. We also present two classification schemes based on multi-class Support Vector Machines (SVMs): (1) Single multi-class SVM and (2) Attribute Bagging-based Ensemble of multi-class SVMs. These classification schemes allow simultaneous learning of the differences between global representations of different classes and the similarities between different shapes or generations within a same make-model class, to further overcome the multiplicity challenges for real-time application. Extensive experiments conducted using our approaches yield superior results for images that were occluded, under low illumination, partial camera views, or even non-frontal views, available in a recently published real-world VMMR dataset. The approaches presented herewith provide a highly accurate VMMR system for real-time applications in realistic environments.

Recently, adding more lanes becomes less and less feasible, which is no longer an applicable solution for the traffic congestion problem due to the increment of vehicles. Using the existing infrastructure more efficiently with better traffic control and management is the realistic solution. An effective traffic management requires the use of monitoring technologies to extract traffic parameters that describe the characteristics of vehicles and their movement on the road. A three-dimension glass fiber-reinforced polymer packaged fiber Bragg grating sensor (3D GFRP-FBG) is introduced for the traffic monitoring system. The proposed sensor network was installed for validation at the Cold Weather Road Research Facility in Minnesota (MnROAD) facility of Minnesota Department of Transportation (MnDOT) in MN. A vehicle classification system based on the proposed sensor network has been validated. The vehicle classification system uses support vector machine (SVM), Neural Network (NN), and K-Nearest Neighbour (KNN) learning algorithms to classify vehicles into categories ranging from small vehicles to combination trucks. The field-testing results from real traffic show that the developed system can accurately estimate the vehicle classifications with 98.5 % of accuracy. Also, the proposed sensor network has been validated for low-speed and high-speed WIM measurements in flexible pavement. Field testing validated that the longitudinal component of the sensor has a measurement accuracy of 86.3% and 89.5% at 5 mph and 45 mph vehicle speed, respectively. A performed parametric study on the stability of the WIM system shows that the loading position is the most significant parameter affecting the WIM measurements accuracy compared to the vehicle speed and pavement temperature. Also the system shows the capability to estimate the location of the loading position to enhance the system accuracy.

No description available.

No description available.

No description available.

For Heavy-duty vehicles (HDVs), the distribution of vehicle miles traveled (VMT) by vehicle type is the most significant parameters for onroad mobile source emissions modeling used in the development of air quality management and regional transportation plans. There are two approaches for the development of the HDV VMT distribution; one approach uses HDV registration data and annual mileage accumulation rates, and another uses HDV VMT counts/observations collected with the FHWA truck classification. For the purpose of emissions modeling, the FHWA truck classes are converted to those used by the MOBILE6.2 emissions rate model by using either the EPA guidance or the National Research Council conversion factors. However, both these approaches have uncertainties in the development of onroad HDV VMT distributions that can lead to large unknowns in the modeled HDV emissions. This dissertation reports a new heavy-duty vehicle visual classification and activity estimation method that minimizes uncertainties in current HDV conversion methods and the vehicle registration based HDV VMT estimation guidance. The HDV visual classification scheme called the X-scheme, which classifies HDV/truck classes by vehicle physical characteristics (the number of axles, gross vehicle weight ratings, tractor-trailer configurations, etc.) converts FHWA truck classes into EPA HDV classes without losing the original resolution of HDV/truck activity and emission characteristics. The new HDV activity estimation method using publicly available HDV activity databases minimizes uncertainties in the vehicle registration based VMT estimation method suggested by EPA. The analysis of emissions impact with the new method indicates that emissions with the EPA HDV VMT estimation guidance are underestimated by 22.9% and 25.0% for oxides of nitrogen and fine particulate matter respectively within the 20-county Atlanta metropolitan area. Because the new heavy-duty vehicle visual classification and activity estimation method has the ability to provide accurate HDV activity and emissions estimates, this method has the potential to significantly influence policymaking processes in regional air quality management and transportation planning. In addition, the ability to estimate link-specific emissions benefits Federal and local agencies in the development of project (microscale), regional (mesoscale), and national (macroscale) level air quality management and transportation plans.

The aim of this thesis is to propose methods for obtaining the additional vehicle parameters from the real-world traffic situations images and including existing information of the license plate and localization of the vehicle in the scene. The task is to use an existing installation of the camera systems based on data obtained from these devices to propose a new method of extraction of other vehicle's parameters. Solutions can be divided into two groups: 1. Methods for obtaining the features and methods of data evaluation which will lead to a vehicle's type identification based on a single image of the vehicle. 2. Methods for obtaining data of the vehicle's shape based on image sequence of passing vehicle.

This project is dedicated to the problem of vehicle classification using inductive loop sensors. We created the dataset that contains more than 11000 labeled inductive loop signatures collected at different times and from different parts of the world. Multiple classification methods and their optimizations were employed to the vehicle classification. Final model that combines K-nearest neighbors and logistic regression achieves 94\% accuracy on classification scheme with 9 classes. The vehicle classifier was implemented in C++.