Guo Ruipeng

Study of Support Vector Regression for the in-process optical measurement of surface roughness using laser scattered images

Results from an optical technique for in-process measurement of surface roughness using laser scattered images are presented. Based on light scattering principle, an experimental system that consists of a collimated laser diode, a screen and a CCD sensor is designed to measure surface roughness. The parameters such as a modified scattering feature, bright points ratio and bright grey ratio are obtained from the scattered images. A machine learning technique called support vector regression (SVR) is developed to determine surface roughness. The three features are chosen as input parameters, and surface roughness is selected as output of the SVR model. Experimental results show that the proposed method is effective for the optical measurement of surface roughness with a satisfactory accuracy. The proposed system combined with a transparent window method can be applied to in-process measurement of the surface quality of a machined component.

A new stereo matching algorithm for UAV autonomous navigation

Autonomous navigation is of great significance for information access and landing of unmanned aerial vehicle (UAV). It is the inevitable trend of UAV development. As an excellent image navigation technology, stereo vision can provide critical information for UAV autonomous navigation. The existing stereo matching algorithms reach a low matching accuracy in amplitude distortion images. For the mismatch problem caused by Census transform, a joint matching algorithm consisting of Census transforms and image color information is proposed in this paper. At the same time, the orthogonal integration method is used to raise the speed according to our theory analysis. First, the initial matching cost is constructed by combining the Census transform and color information. Second, the matching cost is aggregated in an improved adaptive window and we use the orthogonal integration method to raise the speed. Finally, after the optimization and re nement, the disparity map can be acquired. The experimental results demonstrate that our method is of higher accuracy by 40% 50% compared with the Census transform method or the image gray information method. The calculation speed has risen by 3 to 12 times. It is robust to the amplitude distortion image and the method can be used in the UAV autonomous navigation very well.