Mobile Robot Vision Location Based On Improved SIFT Algorithm Research

With the continuous development of the mobile robot technology, and mobile robots'widely application in production and life, there emerges an upsurge of mobile robot research in institutions and universities. The visual orientation of mobile robot is one of the key technologies to achieve the artificial intelligence, and it's also an important research topic in the mobile robots study which has drawn the most attention and challenge.Firstly, provide an overview to the development of the mobile robot and the current researches on technology of mobile robot vision, and reach on positioning technology and mapping. In order to achieve the mobile robot localization on the basis of monocular vision, which needs to extract the features of an environment as visual landmark and map-building, adopting the scale invariant feature transform (SIFT feature) to describe the environment.Secondly, as the disadvantage in time costing of SIFT algorithm makes it difficult to meet the real-time of the mobile robot localization, proposed an improved version of the SIFT algorithm, Simplified the feature descriptor generative process. Through the simulation experiment indicated that the improved version of SIFT have features in scale invariance, rotation invariance, illumination invariance and so on and time complexity of the SIFT algorithm is greatly reduced.Finally, using the EKF-SLAM location model which based on improved SIFT feature points, we derived the motion model and observation model of the mobile robot which can achieve precise positioning of the mobile robot. EKF adopts recursive filtering method and obtains the new state estimation by the recursive equation of the data. For the nonlinear system, EKF is an optimal state estimation in statistical significance. Within the framework of the random process, SLAM achieved the estimation of the mobile robot state and environment landmark state at the same time by way of the probabilistic inference which is considered to be crucial for autonomous mobile robots. The real-time and effectiveness of the method have been verified by the simulation experiment.