Green Vision e Sistemi Embedded

Abstract: The PhD research is mainly focused on hardware and software architecture of embedded systems, battery powered embedded smart cameras, heterogeneous sensor network and their applications in computer vision. It tackles the challenges of the limited resources such as computational power, memory and energy. The work analyzes several issues concerning the containment of energy consumption and the implementation of light-weight computer-vision algorithms and methodologies that aim to increase the energy-efficiency of the embedded smart cameras. We will refer to all these challenging topics with the term Green-Vision. Therefore, the thesis describes some cases of study concerning embedded systems and heterogeneous sensor networks in the context of Green-Vision: 1) The design and development of a low cost measuring system of Veiling luminance based on the use of a CMOS image sensor and on a System on Chip (SoC) fully implemented on a FPGA. The soft processor is used to handle image acquisition and all computational tasks need to compute the Veiling Luminance value. The advantages of this single chip FPGA implementation include the reduction of the hardware requirements, power consumption, and system complexity and flexibility. The problem of the high dynamic range images have been addressed with multiple acquisitions at different exposure times. Vignetting, radial distortion and angular weighting, as required by veiling luminance definition, are handled through a single integer look-up table access. 2) A comparison of the two most advanced algorithms for connected component labeling, highlighting how they perform on a soft core SoC architecture based on FPGA. In particular, a previously developed block based connected components labeling algorithm, optimized with decision tables and decision trees has been tested. The results highlight the importance of caching and data cache sizes for high performance image processing tasks. 3) The Study of methodologies that aim to increase the energy-efficiency and battery-life of an embedded smart camera. These methodologies are based on hardware operations performed at the level of the image sensor. They are exploited to perform object detection and tracking on a battery powered embedded smart camera. The use of these techniques reduces the amount of data that is moved from the image sensor to the main memory at each frame. The better use of the memory resources results in a significant decrease in energy consumption and an increase in battery-life. 41.24% decrease in energy consumption, and 107.2% increase in battery-life. 4) On the field of heterogeneous sensor network and sensor fusion, it has been presented a procedure for the mutual calibration of camera motes and Radio Frequency Identification Devices (RFIDs) for people localization and identification. This topic faces the problem of localizing and identifying objects with the final aim to perform intruder detection in wide open area. The procedure only demands a training phase with a single person moving in the scene holding a RFID tag. The results demonstrate that this calibration is sufficiently accurate to be applied whenever different scenarios, where area of overlap between the field of view of a camera and the field of sense of a (blind) sensor must be efficiently determined. While points 1, 2 and 4 have been studied at ImageLab of DII in collaboration with local companies, the research of point 3 has been carried out in a joined collaboration between the ImageLab and the Smart Vision System Laboratory, University of Nebraska-Lincoln, USA. In conclusion, this thesis has allowed to explore the potentiality of embedded devices for Green Vision by addressing some of the most challenging problems in embedded systems, namely the energy efficiency of devices, the implementation in flexible architectures such as FPGA of highly-demanding algorithms, and the fusion of heterogeneous sensor modalities for high level tasks.

Citation:

Santinelli, Paolo "Green Vision e Sistemi Embedded" 2012

 not available

Paper download:

• Author version: