Dense Matching Algotithm Based On Multiwavelet And Trifocal Tensor

Precision of measurement is decided for precision of stereo matching. If the objectsurface have more detailed decription, high density point cloud must be rebuilt. Featurepoints must match exactly dense. Therefore, actively carrying out dense matchingalgorithm research has important scientific significance that it can improve the precisionof3D information recovery and the precision of visual measurement.This paper studies matching based on the feature and region. The performance ofeach method is validated. Dense matching is studied on this basis, and combined withmultiwavelet, trifocal tensor and disparity growing.GHM multiwavelet coefficient matrix is provided combining with the basic principleof wavelet transform based on its localization of frequency domain decomposition and thetime-frequency window analysis feature. Dense disparity map extraction algorithm basedon multiwavelet is proposed. The influence of image noise on dense matching is solved.Dense matching is obtained. The detail of the image information is retained. Highaccuracy of the depth map is restored. the performance of the algorithm is verified byexperiment. Disparity map reliability of curve is drawn.Properties and calculation method of trifocal tensor are provided based that featurematching is guided on basic matrix. Dense matching algorithm based on disparity growingand trifocal tensor is proposed. Gradient principal direction angle difference histogram ofSIFT point is used for increasing the number of strereo matching point. Trifocal tensor isused for constraint disparity growing. Dense matching of three views is obtained.