双目视觉的原木径级快速检测算法

针对成捆原木自动化检尺中原木端面径级检测的关键问题,采用双目立体视觉及图像分割的原理,完成原木径级的快速三维测量.根据原木的直方图特征,提出基于最大熵阈值分割的区域标识算法,设定动态阈值,实现对原木端面与背景的精确分割.将提取的左右图像中原木端面边缘,借助ORB(Oriented FAST and Rotated BRIEF)特征点检测方法,与极线几何理论相结合完成原木边缘的快速立体匹配,得到三维坐标.此外以成捆原木为检测对象,进行原木边缘图像的最小二乘法椭圆拟合,确定原木端面长、短径参数.实验结果表明:该算法能够在10s内完成原木径级的检测,测量误差在2mm内.

Rapid detection algorithms for log diameter classes based on binocular vision

Aimed at the key problem that automatic detection of log piles diameter classes,by the binocular stereo vision and image segmentation principle,3D information of log-end is determined rapidly.According to the histogram feature of log-end,a region labeling method based on the maximum entropy threshold segmentation is presented,which sets the dynamic threshold to achieve the accurate segmentation of the log-end area and background.Meanwhile,with the help of the ORB feature point detection method,combined the epipolar geometry theory with stereo matching,the 3D coordinates are obtained rapidly.Otherwise taking the log piles as the detection object,the least squares principle is fitted to get the best fitting ellipse and log diameter class parameters of major axis and minor axis.Experiment shows that the proposed algorithms can detect the log diameter classes in 10 s,and the measurement error is in the 2 mm.