Abstract: On account of columnar parts in microscope vision system, an autofocus method of the real-time image characteristics detecting and feature area tracking is presented. Focus evaluation function, focus searching algorithm and focus area selecting is included in the proposed focusing algorithm. The system can achieve autofocus quickly and accurately, and the shortcomings of hill-climbing searching algorithm and local maximum of the searching result can be overcome. Meanwhile, it is important to select an appropriate focus area. The image feature area is used as focus area in the paper. The microscope vision system focuses after detecting the image feature. It can focus accurately at the columnar parts’ edge in the situations of the radius of the columnar parts is shorter than the depth of scope. The standard variance error is 205 μm for coarse focus, and 37 μm for fine focus. Experimental results prove the autofocus method can satisfy the request of the micro-assembly system for focus.

Key words: focus area selecting, focus searching algorithm, autofocus, columnar parts, micro-vision, micro-assembly

摘要： 针对显微视觉中的柱状物体图像的清晰度问题，提出了一种实时检测图像特征并跟踪特征区域进行自动聚焦的方法。该聚焦算法包括聚焦评价函数、聚焦搜索算法和聚焦区域选择。该聚焦搜索算法实现了显微视觉系统下快速准确的聚焦，克服了爬山搜索算法的缺点，有效避免搜索结果陷入局部极大值。显微视觉中对聚焦区域的选择尤为重要，以图像特征区域作为聚焦区域，实时检测该特征区域进行聚焦，实现了在景深小于柱状物体半径的情形下对柱状物体边缘的精确聚焦。粗聚焦后电机位置的标准差为205 μm，精聚焦后电机位置的标准差为37 μm。实验结果表明，该自动聚焦方法能够满足微装配系统对显微视觉的聚焦需求。

关键词: 聚焦区域选择, 聚焦搜索算法, 自动聚焦, 柱状零件, 显微视觉, 微装配