- Ph. D. Dissertation
- RWTH Aachen University
- Institute for Integrated Signal Processing Systems
With recent emerging technological advances, the combined integration of sensing, processing and communication capabilities into city infrastructure enables novel solutions for issues in urban environments. Progressing urbanization imposes a wide variety of challenges. For instance, traffic congestion is one of the central hurdles for maintaining an efficient urban mobility. To alleviate the impact of rising traffic density, a deployment of Smart City infrastructure can enable an effective traffic management and further allow for optimizations of the traffic routing. As one viable option, the required information for establishing such measures can be gathered with distributed sensing approaches. The underlying technological concepts target a seamless integration into the existing infrastructure and cityscape, thereby aiming to provide a comprehensive insight into the real-time traffic flow within the areas of interest. Given the aforementioned background, this thesis investigates techniques for object detection with ultrasonic sensors in urban traffic environments. Specifically, methodologies are presented for two fundamentally different use cases, which cover problems of moving and stationary traffic sensing. The moving traffic scenario considers using a sidefire sensor configuration to measure the cross-sectional traffic flow in a multi-lane scenario. In contrast, the investigations on stationary traffic specifically target parking occupancy sensing in perpendicular parking setups. Ultrasonic sensors are known to exhibit several inherent limitations, and are typically only used as low-cost sensors for first-reflection acoustic distance measurements in short-range scenarios. In this work, the scope of applications is extended by analyzing the complete impulse responses, opening the sensor technology for novel applications, and simultaneously increasing the range of operation. The corresponding system architecture is presented together with a custom platform providing the required capabilities in terms of sensor signal acquisition, digital signal processing and multi-level communication layers. With a modular integration of the embedded platform into street lights, the aspired city infrastructure integration is achieved. As a central challenge from the algorithmic point of view, the sensors operate in a complex time-variant, reflective acoustic environment, and do not provide any directional information. This establishes new requirements for the processing system and object detection techniques. To exploit the information provided by the sensors in the first class of scenarios, consisting of moving traffic, statistical signal processing techniques are combined with density-based clustering techniques for object detection. These algorithms are integrated into an underlying system and evaluation framework to extract additional information from detected objects, and to perform a scenario-based parameter optimization. For the second class of stationary scenarios, realizing parking occupancy sensing, a data-centric approach is selected. This approach integrates machine learning techniques and convolutional neural networks (CNN) with suitable exploration strategies for the neural network topologies. In both scenarios, the proposed techniques are characterized and evaluated with real-world measurement data, while considering the essential trade off of algorithmic performance and computational complexity on the target system.