|
With the deeply development of computer image processing technology, theextraction of information in the3d objects is particularly important in industry,household appliances and other fields. Combining the advantages of varioustechnologies such as laser technology and optoelectronic technology, special3Dmeasurement caused much attention, especially in some measurements areas,such as the important cultural relics, the elastic and plastic material and humanbody. Because it has a non-contact, non-destructive, high resolution, highaccuracy, fast speed, etc. Moreover, this method has some deficiencies. Somehas great accuracy, but speed could not reach the requirements of application;others is the reverse side. Due to this problem, the paper put forward of a newoptical3D measurement method based on neural network, according to thecharacteristics of neural network and the principle of same point phase is same.According to the implementation of the optical3d measurement process,this paper mainly studies the following several aspects: First of all, the presentresearch of optical3D measurement technology has been analyzed. A kind ofimproved methods are put forward according to the shortages of optical3dmeasurement technology; In the second, neural network was introduced into thecamera calibration, according to the neural network’s characteristics ofdescribing the difficult mathematical model and the present commonly usedmethod of camera calibration. Camera calibration method based on RBF neuralnetwork was put forward, due to RBF characteristics of approximating functionwith any degree of accuracy in Comparison of two classic network model. Thelast, According to studied principle and algorithm of stereo vision, and the phaseof same points in the corresponding pictures is equal, Stereo matching methodbased on two-way phase was put forward based on phase matching.Using the method for three dimensional simulation measurements, thismeasurement consists of system calibration, stereo matching and3Dreconstruction. When using RBF network for camera calibration, Networksamples were got by Harris corner method and zhang’s calibration method inorder to high accuracy; In order to improve the matching speed, the phase valuesare got by projected and bidirectional stripe to object in stereo matching. AndCompared with polar and the shortest distance. reconstruction effect of using the method can be seen that has better experimental results.
|