由序列图像进行三维测量的新方法

New method for 3D measurement based on image sequence

目前的三维测量方法都需要专门的测量设备且存在着种种限制,为此提出了一种基于图像序列进行三维测量的新方法。将由数码相机围绕被测物体拍摄的多幅图像导入计算机,利用图像处理知识得到特征的二维信息;采用计算机视觉方法,对特征从射影空间到欧式空间分层逐步重建即可完成三维测量。设计一套特征标志组合,作为辅助测量工具避免了特征匹配难题。确立了一套图像分割与识别策略获得特征标志二维信息,识别率可达到95%以上。采用基于模约束的摄像机分层自标定方法得到特征在欧式空间下的三维信息,并通过多种优化方法减少误差的影响。该方法在硬件上实现简单,对测量条件要求不高。实际试验表明,相对误差可达到1.48%,重投影误差为0.3864像素。

Current methods for 3D measurement have various kinds of restrictions. For example, laser triangulation method has shadow effect while cross-section method destroys the object etc. Moreover, all these methods need special measurement equipments which increase the cost of measurement. A method based on image sequence is proposed. By importing some photos of the test object taken by a digital camera in a computer, the 2D information of the test piece can be obtained through image processing and 3D measurement can be finished by stratified reconstruction of the features from projective space to Euclidean space step by step by means of computer vision. A set of coded targets is designed as an assistant utility to measure the object, which avoids the matching problem. A strategy of image segmentation and recognition is proposed to obtain 2D information from each image. The rate of recognition is beyond 95%. The stratified self-calibration based on modulus constraint is used to get 3D information from Euclidean space. And some optimization algorithms are used to reduce error impact. The implementation of this method is simply done on hardware with little requirements for measuring conditions. Experiments show that the relative error can reaches 1.48% and the re-projection error reaches 0.3864 pixel.