Optimization Method for Model-based Stereo Vision

摘要

The non-model-based binocular stereo vision and the model-based monocular vision are two usual methods used to determine 3D position and orientation of objects. However, the use of these two methods in the long-distance measurement has been limited due to their poor accuracy. If binocular stereo is used instead of monocular vision in the model-based cases, it is possible to, improve the precision because of some redundant information. This paper discussed an optimization strategy to reduce the error of model-based stereo caused by quantification of images. The optimization model concept is explained in detail, and four different objective functions are discussed. In this way, the normal stereo solution can be modified and the solution accuracy can be improved dramatically. The effectiveness of the approach has been confirmed by simulation. The simulated experiments show, by using this method, the measurement error of stereo vision can be reduced by a factor of 50%, and the distribution of the error can also be get better than other common calculation methods.