摘要
棋盘格角点检测技术是计算机视觉的基础,能够用来标定成像系统的参数。现有方法在精度评估方面存在不足,无法得到角点的真值。提出一种计算真值坐标的方法,根据摄像机成像模型,建立反投影模型;再结合镜头畸变模型,计算畸变前后像点之间的映射关系。用该方法得到的真值评估两种角点检测算法,水平方向的平均误差分别为0.06771和1.00788像素,竖直方向的平均误差分别为0.04942和1.00433像素。实验结果表明,该方法可准确评估角点检测精度。根据生成图像的不同,该方法也可用于检测更多类型的角点。
The checkerboard corner detection is an important foundation of computer vision,which can be used to calibrate camera parameters.There are still some shortcomings in the accuracy evaluation as to existing methods for the absence of ground truth.A method to compute ground truth is proposed.We combine lens distortion model and inverse projection model that is constructed according to camera pin hole model to map the pixel coordinates between idea and actual projection.Two common algorithms for corner detections are evaluated by the ground truth acquired by the proposed method,the horizontal average detection errors are 0.06771 and 1.00788 pixel,while the vertical are 0.04942 and 1.00433 pixel.The experiment results demonstrate that the proposed method can be used as ground truth to verify the accuracy of corner detection.Other types of corner also can employ the method for evaluation by modifying the computer generating image.
作者
童飞
TONG Fei(Shanghai X-CHIP Technology Co.Ltd.,Shanghai 201203,China)
出处
《软件导刊》
2021年第5期173-177,共5页
Software Guide
关键词
角点检测
计算机视觉
精度评估
镜头畸变模型
反投影模型
corner detection
computer vision
accuracy evaluation
lens distortion model
inverse projection model
作者简介
童飞(1988-),男,硕士,上海兴芯微电子科技有限公司算法专家,研究方向为机器视觉、图像处理算法。
