📁Machine Learning
👤Zhuxi Li
(Harbin Institute of Technology)
📅Apr 30, 2018
More and more applications like objective detection, objective tracking and facial recognition appear on embedded systems and mobile applications. Convolution neural network (CNN) is the key algorithm to these applications. However, CNNs are computationally and memory intensive which leads to long runtime and high power consumption, making them nearly impossible to deploy. We are working on a CNN software architecture suitable to these applications based on Cyclone V SoC-FPGA hardware platform, which are designed to contain 3 parts:
(1) Compressed and pruned CNN model with less computation and power consumption
(2) CNN accelerator on FPGA based on OpenCL
(3) Matching software and APIs
📁Internet of Things
👤Zhang Jinglong
(Guangdong University of Technology)
📅Apr 30, 2018
At present, IoT devices are in an explosive growth stage. According to IDC statistics, the total investment in the global IoT market in 2016 was 736.9 billion U.S. dollars, and by 2020 this figure will reach 1.2899 trillion U.S. dollars. However, the security issues of IoT devices are also getting worse day by day and there are many devices are vulnerable to security breaches. For example, at the end of 2016, Mirai malware infected a large number of IoT devices and formed a botnet to launch a DDoS attack on the world famous DNS provider Dyn, resulting in the failure of PayPal websites to pay and social websites such as Twitter and Tumblr can not log in. Therefore, the problem of information security can not be effectively solved, which will seriously hinder the popularization and application of Internet equipment. In response to the above problems, this product relies on the FPGA platform to design a SOC (system on chip) that supports multiple cryptographic modules such as the National Security Algorithm SM2 / SM3 / SM4, the International Cryptographic Algorithm AES and the Physical Unclonable Function (PUF) To ensure the information security of IoT devices.
📁Digital Design
👤星雨 陈
(武昌首义学院)
📅May 09, 2018
设计意图:
肖像画是一种描绘具体人物形象的绘画。绘制肖像画并不是人天生的本领,需经过长期的训练,艺术家才能绘制出非常逼真的肖像画,而且人工绘制肖像精力有限,绘画时间长,因此设计一个肖像写真器可以让普通人即使不懂绘画也能得到一幅肖像画,增添了生活的趣味性。
设计思想:
为了能让机器人画出写实的肖像画,首先必须由计算机自动生成人脸的线条画,因此,我们采用基于机器视觉的方法,结合人脸检测和人脸特征提取等技术,设计如何把人脸中的特征轮廓线提取出来,转换为矢量点,作为机器人移动绘画的轨迹。
📁Digital Design
👤嘉锐 连
(重庆大学)
📅Apr 30, 2018
吉他、尤克里里等乐器的初学者总是为不知如何为其乐器调弦而烦恼,同样这类表演者也会在演出曲目间隙花费较长时间为其乐器调弦。不管你的经验水平如何,调弦是一个繁琐且耗费时间的事情,这不仅会打击这类乐器初学者的信心,也会让乐器表演者感到十分麻烦。
本项目旨在设计与实现一款基于FPGA的智能自动调音器,使其能在很大程度上解放吉他、尤克里里这类乐器使用者的双手,在短时间内实现对其乐器的精确调弦。目前,大部分吉他、尤克里里这类乐器的使用者,都使用手动调弦,弦的音调准确只能依靠电子调音器和耳朵确定,多次手动调弦也未必能将弦调到准确音调。与手动调弦不同,使用者只需选择必要参数并将智能自动吉他调音器的旋钮固定在你想要调节的琴弦的头部旋钮上,拨动琴弦,等待数秒钟,就能将跑音的吉他、尤克里里等的琴弦调到正确的音调,相比于手动调弦,实在是太方便快捷了。
📁High Performance Computing
👤Ching-Yi Hsu
(Chung Yuan Christian University)
📅May 06, 2018
隨著智慧型手機的普遍,定位系統亦透過手機大大提升了普遍率,並廣泛地在許多產業中發展,因此有了定位技術。定位技術是行動計算重要的基礎,透過行動載具 (手機) 便能提供位置感知計算的服務以及即時位置相關的資訊給使用者。Kalman Filter (KF) 具有處理模型誤差、測量雜訊以及提升定位系統之精確度之特性,因此本計畫將以 KF 為基礎作為定位的處理機制。本計畫將使用 Field-Programmable Gate Array (FPGA) 實現用於位置追蹤的 KF 演算法,設計與實現 KF 定位演算法將分成幾項部分探討:第一部分為 KF、FPGA 之理論基礎;第二部分為 KF 參數之調整以提高準確度;第三部分為利用 KF 演算法設計的 IC 模型。綜上所述,我們將先了解 KF 之特性,並以此為結果作為未來改善定位系統之前導研究,期以提升定位系統之準確度。定位技術是行動運算中重要的一環,它可以讓移動中的使用者取得即時資訊。本研究計畫將嘗試依環境的不同, 根據演算法算出精確的位置,並將其應用於具有適地性服務 (Location-Based Service,LBS) 應用的平台上。以 FPGA 實現 KF 定位追蹤演算法具有即時定位的特性,以位置資訊為基礎的行動運算可以讓使用者隨時隨地獲得各種資訊各種資源,此研究成果將可延伸應用於醫療、交通、旅遊、娛樂、文化、智慧網路等以 LBS 為基礎的應用系統之相關領域。
📁Digital Design
👤DeZhuang Ma
(Communication Universtiy of China)
📅May 20, 2018
Our 360-degree holographical phantom imaging system based on SoC FPGA use audio processing technology and " pseudo holographic projection" technology to display the real time 3D image of audio spectrum and actual object. On the one hand, SoC Architecture on FPGA can realize the input and process of object image and multichannel audio. On the other hand, 3D holographic projection technology will show the three-dimensional illusion of image so people will intuitively feel the character of music rhythm and the whole picture of the object in front of the camera, creating a striking visual impact.
📁High Performance Computing
👤宇 宋
(长春工业大学)
📅May 27, 2018
我们的目标在于设计一个三维建模系统并安装在无人机系统上以达到能在高空中对大面积地形进行全自动化三维模型构建、形成三维真实场景。
我们提出使用无人机三维建模技术,是因为近年来,消费级旋翼无人机的市场逐渐做了起来,买一个到手就能飞的相机已经是一件非常现实的事情,而在这种条件下,譬如面对各国的位于高山丘陵的自然风景区,使用无人机可以快速得到景区三维鸟瞰图,获得景区的相关信息数据,从而对环境复杂、地形陡峭的自然景区进行完美的规划和设计,达到巧夺天工、浑然天成的理想设计目标。
使用这种技术来对城市进行高空三维建模,构造出的三维城市模型可以使人们摆脱传统的二维平面地图的束缚,使人们对城市景观的现状和设计结果有十分直观的印象,打个比方,经过这种技术处理后,我们能够在虚拟世界中看到大城市的建筑都是立体的,可以看到房子的高度信息,在3D视角下自由移动,从而能够在虚拟世界中欣赏到城市的地貌景观,导航定位更加直观,或是可以对城市的规划设计做一些评估、改进等等。
📁Machine Learning
👤Po-Chun Chien
(National Taiwan University)
📅May 05, 2018
Deep neural network is currently the most powerful machine learning technique. However, it requires high computation effort, especially in MAC (multiply-accumulate) operations, which is not applicable on smaller devices with power constraints.
Stochastic computation requires extremely low cost and power consumption compared to conventional fixed-point arithmetic. However, it comes with random errors and longer latency. In this project, our team is trying to find out a more powerful stochastic computation method to shorten the latency while not sacrificing the accuracy.