Master of Science, The Ohio State University, 2024, Civil Engineering

This thesis explores vehicle classification using returns from LiDAR sensors mounted on a moving instrumented probe vehicle (IPV). The first methodology is a pre-existing scheme that classifies vehicles based on height and length, referred to as the height and length method (HLM). The second is a novel scheme that is developed in this thesis that uses the upper envelope of the vehicle's side view silhouette to classify vehicles, referred to as the shape based classification method (SBCM). The pre-existing HLM was developed using stationary sensors and has already been shown to be robust to data imperfections. The present work demonstrates that HLM also works well when the LiDAR sensors are moving. The main limitation of HLM is that it can only achieve a coarse classification: passenger vehicle, single unit truck, multi-unit truck, or vehicle pulling trailer. The novel SBCM is intended to provide finer gradations among classes. The SBCM collects profiles of the vehicle height from LiDAR returns and normalizes the profiles to 100 points equally distanced along the vehicle (i.e., relative percentage distance from the front to the back of the vehicle). A training set of the empirical LiDAR data collected was used to develop prototype vehicle profiles representative of seven classes: passenger car, SUV, mini-van, pickup truck, van, single unit truck and multi-unit truck. Generally, several sub-class height profiles were developed to capture various vehicle shapes within a class. To develop the profiles, vehicles in the sub-class were chosen from concurrent video imagery as being representative of the sub-class. Then the prototype vehicle height profile for the sub-class was determined by taking the average height across all the representative vehicles at each of the 100 points in the normalized profiles. An eighth classification of vehicles pulling trailers is also employed but does not rely on a prototype height profile. Two variants of the 8-class scheme SBCM were a (open full item for complete abstract)

Committee: Benjamin Coifman (Advisor); Rabi Mishalani (Committee Member); Mark McCord (Committee Member) Subjects: Civil Engineering

PhD, University of Cincinnati, 2015, Engineering and Applied Science: Mechanical Engineering

Recent progress in data collection has enabled the expansion and availability of raw data to take place at an explosive rate. Most of these data, which have been acquired over months and years, are called "time series data". From Prognostics and Health Management (PHM) points of view, it is important to convert raw time series data into useful information quickly and accurately. Therefore, it becomes significant to have comprehensive understanding of data format to support PHM. Most prognostics and health management algorithms hypothesize that the data is a random model drained from a stationary distribution; however, one of the difficult situations for training time series data is a non-stationary environment. Learning in non-stationary environment, also known as learning concept drift, is concerned with interpreting data whose statistical characteristics change over time. Because of the complex intrinsic characteristics of concept drift, learning from such data requires new understandings, techniques, and algorithms to efficiently transform huge amounts of raw data into useful information for PHM usage. This dissertation provides a comprehensive review of three different approaches: physics based model, a data-driven model, and a combination of the two known as a coupled model. These approaches enable the detection of key classification challenges related to the implementation of a prognostics model on time series data. To overcome the classification and concept drift issues with time series data, a novel approach from the coupled model is introduced by applying advanced probabilistic ensemble techniques adapted to the nature of data. For better understanding of the proposed method, its application has been explained in three different case studies. The first case study shows the probabilistic classification of healthy and faulty lithium-ion batteries in a pack using the coupled model. The second case study introduces a method for improving the accuracy of (open full item for complete abstract)

Committee: Jay Lee Ph.D. (Committee Chair); Thomas Richard Huston Ph.D. (Committee Member); J. Kim Ph.D. (Committee Member); David Thompson Ph.D. (Committee Member) Subjects: Engineering

Master of Science, The Ohio State University, 2014, Civil Engineering

Classified vehicle counts are a critical measure for forecasting the health of the roadway infrastructure and for planning future improvements to the transportation network. Balancing the cost of data collection with the fidelity of the measurements, length-based vehicle classification is one of the most common techniques used to collect classified vehicle counts. Typically the length-based vehicle classification process uses a pair of detectors to measure effective vehicle length. The calculation is simple and seems well defined. In particular, most conventional calculations assume that acceleration can be ignored. Unfortunately, at low speeds this assumption is invalid and performance degrades in congestion. As a result of this fact, many operating agencies are reluctant to deploy classification stations on roadways where traffic is frequently congested. This thesis will first demonstrate that small changes in the calculations used in conventional practice can lead to large differences in performance during challenging conditions. This work considers seven different variations of vehicle length calculation, two of which perform much better than the others in congested freeway conditions down to 15 mph- both under theoretical vehicle motions and empirical data analysis. Then, to further improve performance, we evaluate the feasible range of true vehicle lengths that could underlie a given combination of measured length, measured speed, and unobserved acceleration at dual-loop detectors. From this analysis we find that there are small uncertainty zones between length classes where the particular class is ambiguous. For the vehicles falling into the uncertainty zones we assign them to two or more classes- representing all of the feasible true length classes that could have yielded the measured speed and length. The rest of the length-speed plane can be unambiguously assigned to a single class. Finally, using empirical data these advances are shown to perform better (open full item for complete abstract)

Committee: Benjamin Coifman (Advisor); Mard McCord (Committee Member); Philip Viton (Committee Member) Subjects: Civil Engineering

PhD, University of Cincinnati, 2013, Engineering and Applied Science: Civil Engineering

The accurate measurement of vehicle classification is a highly valued factor in traffic operation and management, validations of travel demand models, freight studies, and even emission impact analysis of traffic operation. Inductive loops are increasingly used specifically for traffic monitoring at highway traffic data collection sites. Many studies have proven that the vehicle speed can be estimated accurately by using dual-loop data under free traffic condition, and then vehicle lengths can be estimated accurately. The capability of measuring vehicle lengths makes dual-loop detectors a potential real-time data source for vehicle classification. However, the existing dual-loop length-based vehicle classification model was developed with an assumption that the difference of a vehicle's speed on the first and the second single loop is not significant. Under congested traffic flows, vehicles' speeds change frequently and even fiercely, and the assumption cannot be met any more. The outputs of the existing models have a high error rate under non-free traffic conditions (such as synchronized and stop-and-go congestion states). The errors may be contributed by the complex characteristics of traffic flows under congestion; but quantification of such contributing factors remains unclear. In this study, the dual-loop data and vehicle classification models were evaluated with concurred video ground-truth data. The mechanism of the length-based vehicle classification and relevant traffic flow characteristics were tried to be revealed. In order to obtain the ground-truth vehicle event data, the software VEVID (Vehicle Video-Capture Data Collector) was used to extract high-resolution vehicle trajectory data from the videotapes. This vehicle trajectory data was used to identify the errors and reasons of the vehicle classifications resulted from the existing dual-loop model. Meanwhile, a probe vehicle equipped with a Global Positioning System (GPS) data logger was used t (open full item for complete abstract)

Committee: Heng Wei Ph.D. (Committee Chair); Changjoo Kim Ph.D. (Committee Member); Herbert Bill Ph.D. (Committee Member); Anant Kukreti Ph.D. (Committee Member) Subjects: Civil Engineering

MS, University of Cincinnati, 2010, Engineering and Applied Science: Civil Engineering

The growing congestion problem on Interstates has been identified as a serious problem for accurate data collection from automatic sensors like Inductive loop detectors (ILD). Traffic speed and vehicle classification data are typically collected by dual-loop detectors on freeways. During congestion, measurement of vehicle lengths which is based on detector ON and OFF timestamps (raw loop event data) often lead to misclassification of vehicle data. Accurate detection of raw event data and modified classification algorithm are increasingly important for higher data accuracy needs for agencies such as Advanced Traffic Management Systems (ATMS) and Advanced Traffic Information Systems (ATIS). Vehicle classification algorithm works on the assumption of constant vehicle speed in the detection area. This assumption is violated during congestion which induces errors in to vehicle length estimates leading to more inaccurate vehicle classification data. This paper unlike in preceding works presents a model which is simple enough to be implemented using existing loop detector hardware. This new model assumes vehicle travels with constant acceleration over loop detection area and thus named as ―Constant Acceleration based Vehicle Classification model (CAVC)‖. This model first identifies traffic flow state and later uses Kinematic equations for estimating vehicle length values. Data is collected by videotaping dual loop station and also simultaneously collecting raw loop event data. Ground truth vehicle data is then extracted using Vehicle Video-Capture Data Collector (VEVID) [Wei et al. 2005] from video data. This improved model (CAVC model) is then validated using ground truth classification data and also compared with the results from existing vehicle classification model for different traffic flow states (under specific scenarios).

Committee: Heng Wei PhD (Committee Chair); Anant Kukreti PhD (Committee Member); Changjoo Kim PhD (Committee Member) Subjects: Civil Engineering

Master of Science, The Ohio State University, 2009, Electrical and Computer Engineering

Vehicle detection and classification is important in traffic analysis and management. Various sensing techniques can be used in this field, while most preceding work relies on sensors mounted along the road way, this study develops a mobile platform using a LIDAR equipped probe vehicle to collect ambient traffic data while it drives. A vehicle detection method is developed to extract on-road vehicles from the background. The system employs two LIDAR sensors to measure the speed of the detected vehicles and then their length. A vehicle classification scheme is developed using length and height to sort the vehicles into six pre-defined categories. Ground truth data were generated from a developed GUI interface. Both the vehicle detection algorithm and the vehicle classification algorithm are evaluated by comparing the LIDAR measurement with the ground truth data, with good result.

Committee: Benjamin Coifman (Advisor); Charles Toth (Committee Member) Subjects: Electrical Engineering