基于双目立体视觉的指针式仪表判读算法

Identification and interpretation algorithm for pointer-typemeters based on binocular stereoscopic vision

针对当采集设备的光轴偏离仪表中心时,指针在表盘图像上的投影位置将偏离实际位置,造成读数误差的问题,提出了一种基于双目立体视觉的指针式仪表判读算法。使用标定的双目相机采集指针式仪表图像,通过立体校正使得左右图像在水平方向上对齐。基于阈值分割和Hough变换识别左图指针针尖和仪表盘平面目标点的像素坐标,对针尖位置进行初定位,并利用灰度匹配算法提取右图对应点。利用双目测距原理计算目标点的三维坐标,并构建仪表平面,将针尖垂直投影在该平面上实现指针针尖的重定位,并输出识别读数。实验结果表明:该算法具有较强的稳定性和适用性,相比于基于Hough变换的指针读数识别算法的结果,所提算法降低了约64%的误差。

Aiming at the problem that when the optical axis of the acquisition device deviates from the center of the meter,the projected position of the pointer on the pointer-type meter in the image will deviate from the actual position,making it difficult to accurately recognize the reading,an identification and interpretation algorithm for pointer-type meters based on binocular stereoscopic vision is proposed.Firstly,the image of pointer-type meter is captured by using calibrated binocular camera,and the left and right images are aligned in the horizontal direction by stereoscopic correction.Then,the pixel coordinates of the pointer tip and the target points on the meter panel plane in the left image can be recognized based on threshold segmentation and Hough transform,which determines the initial position of the pointer tip.And similarly,the gray matching algorithm is used to extract the corresponding points in the right image.Finally,the three-dimensional coordinates of the target points are calculated by the distance measurement based on binocular stereoscopic vision and the meter plane can be constructed with the three-dimensional information.The pointer tip is projected vertically on the plane to reposition the pointer tip.The recognized reading can be output.The experimental results show that the algorithm has strong stability and applicability.Compared to the results of pointer reading recognition algorithm based on the Hough transform,this algorithm in this paper reduces the error by approximately 64%.