CONFERENCE PROGRAM

The finalized conference program is now available HERE.

A detailed list of the accepted papers can be found HERE.

Keynote Talk 1

Title: Significant Progress of Analysis of ColorImage Sequences

Speaker: Prof. Ayoub K. Al-Hamadi

Synopsis: The analysis of color image sequences has an essential influence on the current research in the area ofintelligent surveillance, environment perception, driver assistance systems and human computer interaction.In the context of detecting and tracking moving objects in color image sequences, four essentialresearch aspects will be treated in this presentation, which describe the contributions and innovationsfor solving the correspondence problem in complex image measuring situations. To address the problemsof the fluctuation detection and object interactions a data association step is suggested whichincludes data exclusion, data allocation and data administration. Moreover, the reduction of complexityis also an important aspect which we addressed in our work. Based on these aspects and contributions,the work is establishing a fundament for the analysis of image sequences under the influence of noncooperativemeasuring situations and opens the new possibilities for the future research.

Keynote Talk 2

Title: Sensor Networking for Detection: FromDistributed Detection to Energy Savings andMIMO Radar

Speaker: Prof. Rick S. Blum

Synopsis: The focus of the talk is on sensor networking for signal detection. We give a brief review of distributed signal detection which describes some very early work on sensor networking for signal detection. We discuss the important result that sensor likelihood ratio tests are optimum under independence from sensor to sensor. We discuss the more difficult cases of statistically dependent observations and show that some progress can be made in these cases. Next we describe some new work on energy savings for signal detection that shows traditional approaches can be significantly outperformed by ordering the sensor transmissions to send the most informative data first. Finally we discuss a new paradigm called MIMO radar where widely separated multiple transmitters and receivers are employed using either coherent or noncoherent processing. The noncoherent processing allows diversity gains similar to those obtained in communications. The coherent processing allows very high resolution estimation of the position and velocity of objects of interest.

Technical Talk (National Instruments)

Title: Extending Image and Signal Processing to New Application Areas with Real-Time High Performance Computing

Speaker: Dr. Dinesh Nair

Synopsis: Science and industry continue to push the limits on the resources and computing power of single processors available today. Image and signal processing techniques are being used in new applications such as signal Intelligence in the telecom industry; Optical Coherence Tomography (OCT) and DNA sequencing in the medical / life sciences area; precision process control in the semiconductor industry and x-ray based inspection in the electronics industry. These applications are characterized by huge amounts of data to be processed in real-time and stringent demands on processing power. These engineering challenges, from image and signal processing to advanced modeling and simulation, continue to require more and more processing ability.

Traditional high performance computing (HPC) models, where the data is collected first and then calculations are performed off-line using a computer cluster cannot be used to address these applications. Traditional HPC is useful when data is processed a small number of times and real-time feedback is not needed. These new applications require the processing to be done in-line with acquisition of data from real world I/O. For example in an image-guided laser surgery application, the medical treatment system needs to process multi-model images such as CT, PET and MRI, with multi-resolution requirements in real-time. The system uses a model-based predictive control system that involves adaptive feedback for real-time control of a laser probe and online simulations of complex thermal environments that evolve during surgery. This requires a visualization module that allows users to visualize the bioheat transfer process all the way from the tissue level to blood vessels and nanoscale levels. High resolution images of tumors, surrounding tissue, and evolving temperature fields must be digitized, transmitted to a visualization arena, rendered with three-dimensional geometric representations, dynamically meshed, and used to calibrate the bio-heat transfer models in near real-time.

To successfully address the requirements for these applications and to enable continued growth in these areas, we need to fully harness the performance potential of specialized processors such as Field Programmable Gate Arrays (FPGAs), General Purpose Graphics Processing Units (GPUs) and multicore CPUs and address challenges in parallel programming on these and other targets. Real-time high performance computing (RTHPC) combines off-the-shelf technologies to enable the use of massive processing power directly with real-world I/O.  In this talk we will take a look at how a real-time high performance computing approach can be used to address some of the emerging engineering challenges. We will look at the challenges in developing image and signal processing algorithms that fully utilize the heterogeneous computing platform that is available today.  

 Through several examples spanning from signal intelligence, OCT, DNA sequencing to target (laser) tracking, this talk will take your through the various considerations that need to be addressed to map these applications to a single heterogeneous computing platform comprising of FPGAs, multi-core CPUs and GPUs. Although the talk will focus on image and signal processing applications, RTHPC can be used as a building block to solve some of the largest science and engineering challenges in areas such as plasma physics, distributed robotics, smart grids and extremely large telescope control.This talk will also highlight the work being done towards a graphical system design approach that allows a domain expert to work on a single platform and off the shelve hardware to rapidly design and prototype computationally intensive, real-time image and signal processing applications such as 3D imaging and signal intelligence.