红外与激光工程, 2020, 49 (5): 20190553, 网络出版: 2020-09-22  

用于实时目标检测的高速可编程视觉芯片

A high speed programmable vision chip for real-time object detection
李鸿龙 1,2,*杨杰 1,2,3张忠星 1,2罗迁 1,2于双铭 1,2刘力源 1,2吴南健 1,2,4
作者单位
1 中国科学院半导体研究所 超晶格国家重点实验室,北京 100083
2 中国科学院大学 材料与光电研究中心,北京 100049
3 西湖大学 工学院,浙江 杭州 310000
4 中国科学院脑科学与智能技术卓越创新中心,北京 100083
摘要
视觉芯片是一种高速、低功耗的智能视觉处理系统芯片,在生产生活中有广阔的应用前景。文中提出了一种新型的可编程视觉芯片架构,该架构的设计考虑了传统计算机视觉算法和卷积神经网络的运算特点,使其能够同时高效地支持这两类算法。该视觉芯片集成了可编程的多层次并行处理阵列、高速数据传输通路和系统控制模块,并采用65 nm标准CMOS工艺制程流片。测试结果表明:视觉芯片在200 MHz系统时钟下达到413GOPS的峰值运算性能,能够高效地完成包括完成人脸识别、目标检测等多种计算机视觉和人工智能算法。该视觉芯片在可编程度、运算性能以及能耗效率等方面都大大超越了其他视觉芯片。
Abstract
Vision chip is a high-speed, low-power intelligent vision processing system on a chip (SoC), it can be applied to many different fields of production and daily life. A new programmable vision chip architecture was proposed which took into account the computational features of traditional computer vision algorithms and convolutional neural networks, it enabled the architecture to support the two types of algorithms simultaneously. The vision chip integrated multi-level parallel programmable processing array, high-speed data transmission path and system control module, it was fabricated in a 65 nm standard CMOS process. The experimental results show that the vision chip achieves 413GOPS peak performance at 200 MHz system clock and can perform various computer vision and artificial intelligence algorithms including face recognition and target detection efficiently. The proposed work shows higher system performance in terms of programmability, performance and energy efficiency when compared with other state-of-the-art vision chips.

李鸿龙, 杨杰, 张忠星, 罗迁, 于双铭, 刘力源, 吴南健. 用于实时目标检测的高速可编程视觉芯片[J]. 红外与激光工程, 2020, 49(5): 20190553. Li Honglong, Yang Jie, Zhang Zhongxing, Luo Qian, Yu Shuangming, Liu Liyuan, Wu Nanjian. A high speed programmable vision chip for real-time object detection[J]. Infrared and Laser Engineering, 2020, 49(5): 20190553.

引用该论文: TXT   |   EndNote

相关论文

加载中...

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!