Abstract: In the light inspection machine system for oral liquid impurity detection, due to slight vibration of oral liquid bottle caused by sudden halt of crawler belt used for bottle spinning, in the sequential two pictures of oral liquid bottle captured from high-speed industry camera, pixels sharing the same partial positions cannot be overlapped, which may lead to incorrect result after differential operation of the two pictures. For impurity in oral liquid is very small about micrometer level, mechanical dithering or dirt on the bottle of oral liquid shall pose influence on inspection accuracy because of aforementioned reason. This paper will harness SIFT algorithm to register the sequential two pictures. SIFT algorithm is suitable for high accuracy oral liquid impurity detection system on account of its stability. It improves SITF a little in the aim of adapting to slight vibration range which will generate satisfactory result. In the real process of detection, the algorithm is of high stability and high accuracy.

Key words: light inspection machine for oral liquid, high-accuracy impurity detection, Scale Invariant Feature Transform（SIFT）, position registration

摘要： 在口服液灯检机杂质检测系统中，口服液瓶体由于履带搓瓶的急停会有轻微的抖动，造成高速工业摄像机拍摄的前后两帧口服液瓶体图像中位于相同空间位置的像素无法重合在一起，导致前后两帧图像做差分结果出现错误。由于口服液中的杂质很小，一般会达到微米级别，因此机械的扰动以及口服液瓶体上的污点都有可能因位置偏差对检测结果造成影响。采用尺度不变特征检测（SIFT）对系统采集的前后两帧图像进行位置配准。SIFT算法稳定性精度很高，适用于高精度口服液杂质检测系统。基于抖动幅度微弱，对该算法进行了一定的改进与简化，以获得最佳配准结果。在实际检测过程中算法稳定，检测结果准确率很高。

关键词: 口服液灯检机, 高精度杂质检测, 尺度不变特征检测（SIFT）, 位置配准