*********************************
There is now a CONTENT FREEZE for Mercury while we switch to a new platform. It began on Friday, March 10 at 6pm and will end on Wednesday, March 15 at noon. No new content can be created during this time, but all material in the system as of the beginning of the freeze will be migrated to the new platform, including users and groups. Functionally the new site is identical to the old one. webteam@gatech.edu
*********************************
Title: Resource-aware and Robust Image Processing for Intelligent Sensor Systems
Committee:
Dr. Saibal Mukhopdhyay, ECE, Chair , Advisor
Dr. Sudhakar Yalamanchili, ECE
Dr. Arijit Raychowdhury, ECE
Dr. Tushar Krishna, ECE
Dr. Matthai Philipose, U of Washington
Abstract:
The objective of this research is to design resource-aware and robust image processing algorithms, system architecture, and hardware implementation for intelligent image sensor systems in the Internet-of-Things (IoT) environment.
The research designs an wireless image sensor system with low-overhead pre-processing, which is integrated with a reconfigurable energy-harvesting image sensor array to implement a self-powered image sensor system. For reliable delivery of region-of-interest (ROI) under dynamic environment, the research explores the design of low-power moving object detection with enhanced noise robustness. The system energy is further optimized by a low-power ROI-based coding scheme, whose parameters are dynamically controlled by a low-power rate controller to minimize required buffer size.
To enable machine learning based intelligent image processing at the IoT edge devices, the research proposes resource-efficient deep neural networks. The storage demand is reduced by compressing the neural network weights with an adaptive image encoding algorithm, and the computation demand is optimized by mapping the entire network parameters and operations into the frequency domain. To further improve the energy-efficiency and throughput of the edge device, the research explores inference partitioning of a DNN between the edge and the host platforms.
