| In the applications of target recognition and target location et al, it is one of themost critical processes to affine invariant features extraction that immediate impacts thefinal results of the relevant applications. Supported by the National Nature ScienceFoundation of China (608712136), named “Research on ranging system and keytechnology based on two frames in image sequence from monocular vision and theimage direction” and the Natural Science Basic Research Plan in Shaanxi Province ofChina (2011JM8002), named “Research on the virtual circle characteristic and itsapplication in target passive ranging”, we conduct the research in target’s rotationinvariant features and its applications, and extend the rotation invariant circle feature oftargets to the virtual circle feature.A review was given on the target’s rotation invariant feature extraction technologyand the characteristics about each type of invariant feature. Then a concept of virtualcircle based on multipoint features was proposed as per the inherent rotation invarianceof a circular target. In this theme,5different kinds of virtual circles were constructed byusing3pairs of matching points in adjacent frames in the image sequence, andcompared their performance. Compared with the existing methods, this kind of featuresare easier to extract and need less constraint condition on the matching points. It isdemonstrated by our simulations that the diameter feature of the virtual circle based onthe extended equilateral triangle is a preferable depth-relate line segment feature. Underthe condition of the variation rang of the inclination angle of target relative to camera isbetween [-10°,10°], the target ranging error by using this line segment feature is about±3%. Furthermore, we achieved relative pose measurement between adjacent imageframes using4matching points. Experimental results indicate that the measure error isless than2.5%under a pixel noise. Furthermore, this method has preferable noiseimmunity and robustness. |
PDF Full Text Request