一种嵌入式智能视觉系统的设计

Design of Embedded Intelligent Vision System

针对目前嵌入式视觉系统的图像识别算法复杂、鲁棒性差等问题,提出了一种嵌入式智能视觉系统。该系统以进阶精简指令集机器Cortex-A53为核心搭建硬件平台,并搭载Linux嵌入式操作系统。基于V4L2接口设计图像采集程序,完成图像的采集及存储。通过移植学习框架,采用Inception-V3神经网络模型实现图像的智能化识别。实验结果表明:该系统具有良好的采集效果,对不同物体和同一物体在不同姿态下的识别准确度均可达100%,满足智能视觉系统的设计要求。

In view of the high complexity of image recognition algorithm and poor robustness of present embedded vision system,the embedded intelligent vision system was proposed.The advanced RISC machine(ARM)Cortex-A53(Allwinner H5)was taken as the core of the system to buildthe hardware platform,and the system was powered by Linux embedded operating system.The image acquisition program was designed based on the V4L2 interface to realize the image collection and storage.The neural network model was adopted to achieve the intelligent recognition of the image by transplanting the TensorFlow learning framework.The experimental results showed that the recognition of different objects and the same object under different postures can reach 100%,which meets the design requirements of intelligent vision system.