基于快速模板匹配的智能视觉传感器设计*

摘要

A design of a smart digital visual sensor for inline visual inspection system for the white body of cars is reported. Based on ARM 6410 processor,the system integrates the functionalities of data collection,2D pixel-level locating,communication as well as controlling,and it aims at solving the problem of poor signal processing capability and bad immunity to interference in traditional visual sensors. A fast and stable pixel-level template matching algorithm based on Linux system is designed,which uses NNPROD as similarity measurement and adopts a multi-resolution strategy. In the matching algorithm, a new 2D summation method called “Running Sum with Complementing Lost Pixels” is proposed,which optimizes the calculation of the cross-correlation coefficient. With a comprehensive fault-tolerant mechanism in addition to the matching algorithm,the visual sensor has been proved ex-perimentally to run the pixel-level locating function with high speed, high precision and high robustness, and to fulfill the task of front-end data collecting and processing in the inspection system.%针对传统视觉传感器信号处理能力差、模拟信号输出易受干扰等问题，设计了基于ARM 6410处理器的智能化数字视觉传感器，应用于白车身在线视觉检测系统，完成数据采集、像素级二维定位和通讯控制工作。在Linux平台上，研究了一种快速稳定的像素级模板匹配算法，以去均值归一化灰度互相关法为相似性测度，多分辨率金字塔匹配法为搜索策略，使用“滚动相加补全像素法冶的二维求和新方法来优化计算过程，并进行了全局冗错设计，确保传感器工作稳定性。现场实验表明：采用该匹配算法和冗错机制使得视觉传感器定位功能速度快、精度高、鲁棒性好，能较好完成白车身视觉检测系统前端数据采集和处理任务。