【摘要】 针对当前机器视觉检测领域中单目图像传感器无法对微小医疗器械实现高精度检测的缺点,提出一种基于双目图像传感器的微小零件检测方法。从系统软硬件方面对拼接的精度进行了研究,提出了一种改进后的Harris算法用于特征点精确提取:首先对图像进行形态学处理,然后提取特征点,克服了Harris算子对噪声较敏感、易造成角点位置偏移、失去角点信息等缺点;利用归一化互相关函数对提取的特征点匹配,并用RANSAC算法对误匹配点对进行剔除最终实现亚像素级检测。基于图像拼接的双目机器视觉高精度检测方法,提高了图像分辨率,摆脱了摄像机像素对检测精度的制约,实现微米级检测。该技术同时适用于对超出摄像机视野范围的大尺寸医疗器械检测。更多还原

【Abstract】 Aim at the defect of monocular image sensor which cannot achieve high-precision detection of small medical instruments in today’s machine vision field,a new detection method which based on binocular image sensors is proposed.The factors which can affect the splicing accuracy are discussed from hardware and software aspects,then a new algorithm of improving-Harris using for precise extraction of feature points can be tabled:first,images are processed by morphological method,and then extract feature points.This algorithm overcomes much defects of Harris operator,for example,more sensitive to noise,could easily lead to position offset of corners,loss information of corners,etc.;using normalized cross-correlation function to match the feature points extracted by morphological-Harris,eliminating false matching points by RANSAC,and ultimately achieves sub-pixel detection.High-precision machine vision detecting methods based on image mosaic of binocular cameras can improve the image resolution,and achieve micron level detection without the constraint of detection accuracy by camera pixels.The technology also applies to the large-size medical device testing which beyond the camera field of view.更多还原