报告题目：Using FPGAs for parallel-pipeline processing of video information streams

演讲摘要：In the modern world, there are many devices that require local processing of big information stream. Such devices are LCD and AMOLED panels, which are the basis of modern televisions and displays. FPGAs and specialized integrated circuits (ASICs) are widely used for local processing of big information streams. Video information stream processing requires performing a huge amount of calculations in real time when they are displayed on LCD and AMOLED panels. For example, when displaying an RGB video stream on an RGBW LCD or RGBW AMOLED panel, a color conversion is required, while it is necessary to constantly evaluate the converted video stream to minimize color rendering errors. Another example of video information stream processing is related to the maximum current consumption of AMOLED panels limitation due to adaptive limitation of their total luminosity maximum. The luminosity uniformity of AMOLED panels compensation by the software includes additional decompression algorithms for compressed images performed simultaneously with the processing of the main video stream. Evaluation of the parameters displayed by the AMOLED video data panel allows changing its settings and pre-process the displayed video data at the flow rate, thus improving its characteristics. The use of FPGAs for parallel-conveyor processing of the video information stream allows changing the power consumption not only of the processing device, but also of the LCD and AMOLED panels themselves, as well as to improve their characteristics, such as luminosity uniformity, maximum brightness, contrast, lifetime, cost, etc. At the same time, all calculations are performed at the speed of the video data stream, without frame loss, i.e. in real time.

讲者简介：俄罗斯自然科学院院士、俄罗斯西南州立大学高级研究员，新一代信息技术领域的科学家，2005-2013年作为三星电子集团首席工程师主导高端显示面板驱动芯片研发，自2014年至今，在俄罗斯西南州立大学科学的研究所工作。主要是从事基于人工智能的图像处理、显示驱动芯片设计等应用研究。提出了基于人工智能的三级神经模糊模型的立体重建算法，首次实现基于人工智能的立体重建算法在FPGA（现场可编程门阵列）上工程实现，工程化后的算法性能达到了国际领先水平。2018年，开发了一种基于人工智能模糊逻辑构建深度图的立体视觉系统；2019年提出人工智能软算子深度图及算法；2021年，提出用于绘制深度图的改进Canny边缘滤波器；2023年，发表基于人工智能三级模糊模型的深度图生成方法，并基于FPGA进行了算法验证，解决了论文算法到工程化应用的鸿沟。拥有10多项人工智能图像处理芯片装置发明专利，具备从人工智能算法到芯片设计实现的工程化能力。研究基于AMOLED（有源矩阵有机发光二极体）和LCD（液晶显示面板）的人工智能显示驱动以及节能方案的芯片技术，建立以FPGA可编程IC器件为基础的芯片架构设计验证方案，提高了AMOLED和LCD的显示饱和度，同时降低了电力消耗。 基于上述的研究成果申请了俄罗斯、美国、加拿大等多国工53项发明专利，在当地权威期刊累计发表论文102篇。