Research on feature point localization technology based on binocular vision

AGV navigation indoor mobile object positioning, tracking and service applications in all aspects of life plays a vital role, the market has a more in-depth demand for indoor positioning accuracy and continuity. In this paper, the binocular stereo model with parallel optical axis is used to realize the transformation from two-dimensional coordinates to three-dimensional coordinates, and the feature point detection algorithm and matching algorithm are written for the designed feature points to get the three-dimensional coordinates of the feature points, which can realize the functions of label recognition and position detection. For the actual experimental equipment accuracy, experimental equipment debugging parameters, experimental environment and other factors, the data calculated by the algorithm and the real three-dimensional position of the feature point inevitably has deviation. Therefore, this paper also uses the Kalman filter algorithm to estimate and predict the position information of the feature points extracted based on binocular vision, design related experiments to verify the effect of the above algorithms, analyze the effect of Kalman filter from a three-dimensional point of view, and analyze the three-dimensional data of the origin of the label coordinate system using the Kalman filter for estimation of Kalman filtering algorithm for processing so that the data is relatively smooth, and further improves the localization accuracy, and lays the foundation for the subsequent application of the AGV’s navigation.