Generative Signal Processing and Data Fusion for Sensor Networks

点击翻译按钮获取中文摘要

The recent trends in analytics and interconnectivity between internet of things (IoT) and sensors is transforming industries such as personalized healthcare, agriculture, manufacturing, energy, and surveillance to process data autonomously to gather insight and make informed decisions with minimal delay. Market studies suggest the number of connected sensors is expected to reach 20.8 billion by 2020, and $6 trillion investment is predicted for the sensing market over the next 5 years. Although there exists development in sensor technologies aimed at increasing resolution and efficiency, cost and physical barriers may limit the practical sensor deployment. Due to sensing limitations, phenomena frequently cannot be observed completely due to various factors including sensor failures, sampling or resolution, interference, occlusions from objects, or visibility issues from weather or night-time. These missing information in space, time, and sensor modalities are barriers to real-time continuous sensing. This motivates the need for developing signal processing tools for the sensing picture without being limited by physical sensing constraints. We propose a generative signal processing and data fusion framework to push the boundaries of sensing. Our goal is to develop a unified set of technologies for three fundamental aspects of generative sensing and processing. We propose theoretical approaches and models to find solutions to missing signals, objects, and tracks. We will investigate algorithms for augmenting detection, tracking, and recognition processing tasks with generative signal processing and combine these processes. Finally, we propose an artificial sensor concept and data fusion approaches that will bridge the gap between artificial and real physical sensing. The research results in a framework to generate, process data at desired times, spatial locations, and selected sensor modalities. This will be publicly beneficial in sensing, as personalization and approaches will be developed to acquire desired information in cases where it may be currently expensive or limited. We verify applications in remote sensing, urban and pipeline monitoring with academic and industrial collaborators to drive future commercial technologies, develop next generation sensing for phenomena which cannot be sensed by conventional approaches. Benefits to Canada include a general approach to make eventual trends in continuous sensing more cost-effective, derived algorithms for applications like pipeline, acoustic monitoring for public safety, maritime or arctic monitoring. The research will enhance robustness of decision making in scenarios of sensing failure, to complement current physical sensors and enable continuous sensing.