New approach for 3D shape measurement of complex objects by combining color-coded fringe and neural networks
A new approach for three-dimensional (3D) shape measurement based on color-coded fringe and neural networks is proposed. By applying the phase-shift technique to fringe projection, point clouds with high spatial resolution and limited accuracy can be generated. Stereo-pair images obtained from two cameras can be used to compute 3D world coordinates of a point using traditional active triangulation approach in which camera calibration is crucial. The general camera model is often nonlinear and requires good initial estimates to converge to a solution. Neural network is a well-known approach to approximate a nonlinear system without an explicit physical model, so in this work it is used to train the stereo vision application system to calculating 3D world coordinates so that the camera calibration can be eliminated. The training set for our neural network consists of a variety of stereo-pair images and corresponding 3D world coordinates. In the 3D shape measurement system where neural network is used the picture elements correspondence in image is essential. Because the cameras interior and exterior orientation parameters are not estimated by camera calibration, epipolar condition is unable to be calculated. Therefore the general one-dimensional phase-shift technique can not solve the correspondence problem. In this paper the picture elements correspondence problem is solved by using projected color-coded fringes with different orientations. Once the high accurate correspond points are decided, high precision dense 3D points cloud can be calculated by the well trained net.The obvious advantage of this approach is that high spatial resolution can be obtained by the phase-shift technique and high accuracy 3D object point coordinates are achieved by the well trained net which is not dependent on the camera model and will work for any type of camera. Some experiments can verify the performance of this method.
3D shape measurement color-coded fringe neural networks correspondence problem.
Dahui Qin Zhongwei Li Congjun Wang Yusheng Shi
State Key Laboratory of Material Processing and Die & Mould Technology,Huazhong University of Science and Technology, Wuhan, 430074, China
国际会议
北京
英文
237-242
2008-11-06（万方平台首次上网日期，不代表论文的发表时间）