<p> Due to continuously increasing traffic on highways and roads, the congestion level on the nation’s roadways is spiraling out of control. Current implementations for the traffic detection used in the United States utilize inductive loop technologies placed during road construction, or they are sawed and placed in after. These inductive-loop traffic detectors are primarily used for the detection of vehicles; however, more advanced systems can classify vehicles in addition to detection as well as save large amounts of power. To improve the traffic system and make it more advanced, in this thesis, an in-node microprocessor-based vehicle classification approach to analyze and determine the types of vehicles passing over a 3-axis magnetometer sensor is proposed. This approach for vehicle classification utilizes the J48 classification algorithm implemented in Waikato Environment for Knowledge Analysis (WEKA), a machine learning software suite, and the logistic regression model implemented in MATLAB (MATrix LABoratory) as well. The J48 is a Quinlan's C4.5 algorithm, an extension of the decision tree machine learning based on the ID3 (Iterative Dichotomiser 3) algorithm. The decision tree model is generated from a set of features extracted from the data of the vehicles passing over the 3-axis sensor. The features are attributes provided with correct classifications to the J48 training algorithm to generate a decision tree model with varying degrees of classification rates based on cross-validation. Ideally, using fewer attributes to generate the model allows for the highest computational efficiency due to fewer features needed to be calculated while minimizing the tree with fewer branches. The generated tree model can then be easily implemented using nested if-loops in any language on a multitude of microprocessors. In addition, setting an adaptive baseline to negate the effects of the background magnetic field allows reuse of the same tree model in multiple environments. In addition to the J48, the binary logistic regression model is also used here to estimate the probability of a binary response (0 or 1) based on one or more predictors (or independent) variables (features). The reason the binary logistic model is used in this thesis is because only two classes of the vehicles are being analyzed here. A Sedan is taken as logic 0, and a Sports Utility Vehicle (SUV) is taken as logic 1; both of these vehicle types are the main concern as these two types are quite difficult to differentiate from their magnetic signature.</p><p> The logistic regression measures the relationship between the categorical dependent variable and one or more independent variables by estimating the probabilities using a logistic function. The logistic regression is a special case of the Generalized Linear Model (GLM) and is based on the relationship between dependent (either Sedan or SUV) and independent variables (features). This model predicts the probability of a car being a Sedan or an SUV. If the probability of a car exceeds the required amount, it classifies the car type as either a Sedan or an SUV. The result of this experiment shows that the vehicle classification system is effective and efficient.</p>
| Identifer | oai:union.ndltd.org:PROQUEST/oai:pqdtoai.proquest.com:10263671 |
| Date | 10 May 2017 |
| Creators | Patel, Kuntal G. |
| Publisher | California State University, Long Beach |
| Source Sets | ProQuest.com |
| Language | English |
| Detected Language | English |
| Type | thesis |
Page generated in 0.0016 seconds