UC Berkeley

Intelligent Transportation Systems (ITS) are cost-effective measures to manage congestion due to increasing demand by improving the efficiency of existing transportation infrastructure. Traffic detection and surveillance play a pivotal role in deploying these technologies in the field. This dissertation continues the work that has been done in recent years in relation to the use of wireless magnetic sensor networks in transportation systems. As part of the effort to improve vehicle detection system technologies so that better management strategies can be implemented in the field, the work presented here focuses on advancing the use of wireless magnetic sensors in Intelligent Transportation Systems. This dissertation addresses improvements in algorithmic tools that advance the use of wireless magnetic sensors for both freeways and arterials. The applications addressed here include on-ramp queue estimation, arterial link vehicle-count, travel time estimation on heavily congested arterial streets, travel time and link vehicle-count in freeways, truck re-identification along long freeway segments, as well as cost-effective vehicle classification. The overall goal of this dissertation is to advance the use of these basic detection technologies to roles that extend beyond basic vehicle detection.

A vehicle re-identification system, which relies on matching vehicle signatures from wireless magnetic sensors is modified to improve its performance for stop-and-go traffic conditions and is extended so that it can be used for truck re-identification along long freeway segments. The modifications to the algorithm address problems observed when vehicles stop or accelerate/decelerate as they go through the sensors. The modified system was tested to ensure that it overcame the deficiencies imposed by the original system. The extension of the vehicle re-identification system, presented as the iterative vehicle re-identification system, addresses traffic dynamics observed when vehicles travel along long road segments, in particular, vehicle overtaking. The system was tested extensively to ensure that it can be deployed for truck re-identification along long freeway segments, e.g., in between weigh-in-motion (WIM) stations.

A link vehicle-count and a travel time estimator based on flow-measurements and vehicle re-identification data were studied at a freeway on-ramp, arterial segments as well as at freeway segments. The results show that the estimators are reliable and accurate, and are suitable for real-time traffic responsive management strategies that require precise link vehicle-count and/or vehicle travel time information, such as ramp metering, speed control and traffic intersection control.

Vehicle classification, which utilizes a single wireless magnetic sensor installed in the middle of a freeway lane is also presented. The approach uses a two stage binary support vector machine (SVM) classifier based on features extracted from vehicle signatures. This is a cost effective classification system that uses a small subset of data efficiently extracted from the magnetic signal measured by the sensor. The results showed that vehicles can be reliably and accurately classified into passenger vehicles and trucks, and once trucks are extracted, this group can be further divided, with lower accuracy and consistency, into two groups: small trucks and large trucks.

Finally, this dissertation presents a systematic tool for tuning vehicle re-identification parameters and evaluating performance. This tool uses different plots, metrics and algorithms to evaluate the output of the vehicle re-identification algorithm as well as estimates based on it, i.e., link vehicle-count and vehicle travel time.