Intelligent wireless sensor network based vehicle detection and classification

The purpose of my research in this dissertation is to study issues related to vehicle detection and classification in WSNs, and to provide lessons and solutions by analysis, experiment, and simulations. First, we define the problem by defining the WSN and design goals in our research and identifying the key issues. Next, we present a comprehensive literature review on related works. After that, We introduce the platform and tools we used in experiments and simulation: the Berkeley Mica2 motes and its sensing board as well as TinyOS software.; For vehicle detection, we propose and evaluate a discrete emulation of CFAR detector for mote platform and found its error rate very close to an ideal constant false alarm rate (CFAR) detector. For vehicle classification, we analyze the performance of three common classifiers under two processing schemes using empirical data. Principal component analysis (PCA) is shown to be the most promising one for vehicle classification in WSN among all three. We propose a heuristic to increase the efficiency of PCA signatures. For system design of WSNs, we develop and evaluate several protocols and modules for reliable data transmission, synchronized operations, and collaborative processing, respectively. We profile a mote's energy consumption and latency. Based on those profiles, we study the efficiency of various signal processing schemes and found that their efficiency heavily depends on the network size. By extensive simulations, we study the impacts of resource allocation acid traffic intensity on sensor nodes' lifetime. We found that a sensor node's lifetime is robust to the former while very sensitive to the latter.