激光测距仪镜头感光元件的快速识别与定位算法研究

在激光测距仪镜头组件的自动化装配中,快速、精准识别感光元件雪崩光电二极管(APD)的定位坐标,对于完成镜头焦点的精确对位以及提高图像处理效率至关重要。针对APD的图像识别与定位过程,本文提出一套基于机器视觉的高效、高精度的图像处理算法。首先在粗定位图像处理阶段中,为提高运算效率,利用抗干扰性强的高斯金字塔搜索归一化互相关匹配(NCC)算法,对图像中的APD进行粗定位。在边缘检测中,采用Otsu算法自适应地根据梯度图像变化生成高低阈值,避免了传统Canny算法的手动设置高低阈值的难题。在目标轮廓提取阶段采用连通域标记法,过滤掉孤立的像素点和非目标区域像素点,保证了下一步的轮廓拟合精度。在最后的轮廓拟合精定位阶段中,通过对两种拟合算法比较过程中,确定最小二乘法圆拟合亚像素定位算法进行APD轮廓拟合,可以保证效率和定位精度,实验结果表明整个图像处理系统用时596 ms、定位精度0.4 pixel,相对误差为0.64%,实现了APD图像快速、精准定位的过程,提高了定位精度和效率。

Research on Fast Recognition and Location Algorithm of Photosensitive Component in the Laser Rangefinder Lens

In the automatic assembly of the components of the laser rangefinder lens,it is crucial to get the location of avalanche photodiode(APD)quickly and accurately for the precise alignment of lens focus and improving the efficiency of image processing.Considering the APD image recognition and location process,a highly efficient and high precision image processing algorithm based on machine vision is proposed in this paper.Initially,in the approximate location stage,to improve the computing efficiency,a normalized cross correlation algorithm with a strong anti-interference ability based on thepyramid search algorithm is used to approximately locate the APD in the image.In the edge detection stage,the Otsu algorithm is used to adaptively generate the high and low thresholds according to the gradient image,which avoids artificial setting of high and low thresholds in the traditional Canny algorithm.In the target extraction stage,the contour fitting accuracy uses the connected component analys is labeling,which can filter out isolated noise and non-target pixel points.In the precise location stage,the two fitting algorithms are compared,and circle fitting based on the least square method is used to fit the APD contour and to ensure the efficiency and positioning accuracy,which is in sub-pixel levels.The experimental results demonstrate that the system takes 596 ms to process the image,the positioning accuracy is approximately 0.4 pixel,and the relative error is 0.56%.Thus,the system achieves recognition and location of the APD and improves the positioning accuracy and efficiency.