Present position: freelance worker
|Thesis title:||Project and development of a configurable hardware system based on DSP, FPGA and PC, at kernel and user levels, to process and transmit digital images|
|Research area:||Microelectronics and Emerging Technologies|
In digital electronic systems for audio and video processing is fundamental to treat great amount of data in real time; processing is about information compression, patter recognition and data fusion.
Hardware needed to process information is based on spatial computing devices (FPGA) and temporal computing devices (DSP). Due to different architectural and functional way these devices work, is needed to divide algorithms to take advantage of each hardware to enhance performance at maximum rate.
An intelligent video system, with shape recognition functions able to stream MPEG4 video over network was ideated and realized. Electronics are based on a network processor whose purpose is to stream MPEG4 video and to provide an interface to users through a web server; this processor is also used to execute part of shape detection algorithms that are distributed over three different processors. Innovative object detection algorithms used to identify user defined interesting objects, written with limited matrices computing and a new vectorial analysis, suitable for embedded systems, was realized. This algorithm takes advantage of spatial computing, where basics image processing are implemented, introducing just few microseconds image delay, and sequential processing, where is implemented vectorial object detection and recognition section. To monitor all electronics and to debug all circuits a new real-time operating system for embedded microcontrollers was realized. This operating system has a pre-emptive kernel with full support for common buses and a complete VT-102 console to interact with users. To complete this project and make it homogeneous, a multi-platform, user and kernel image processing framework for PC and embedded systems was realized. This framework uses DirectX technology advantages on Windows based machines and a custom extension to V4L2 on Linux based system.
The main goal of this research is to demonstrate that real-time advanced image processing can be done also on limited resources embedded hardware using only commercially available devices if hardware is correctly planned and software is designed to run on low performance processors with adequate algorithms.
Results obtained from this research are very promising, a new vectorial image analysis system was developed, an innovative image framework and a real-time operating system was developed, everything running on an embedded dedicated hardware platform; vectorial analysis system it is currently used to recognize shapes into images and can be further enhanced to implement advanced features, as an example to implement a movement prediction system suitable for compression algorithms.