Research Of Key Technologies In Vision-Guided Material Handling AGV

Manufacturing is a basic, frontier, pillar, and strategic industry of a country. With the emergence and application of lean production, flexible manufacturing, intelligent manufacturing, computer integrated manufacturing, green manufacturing, JIT and other advanced manufacturing models, manufacturing enterprises call for more advanced manufacturing equipment and material handling systems. Automatic Guided Vehicle(AGV), one of the key equipment of modern industrial logistic systems, sees an important research orientation on how to further enhance its intelligence, flexibility at present and in the future. Reviewing the domestic and foreign situations of material handling AGV, and application status of vision-guided AGV, and key technologies of machine vision, the present study attempts to deeply explore the target tracking and recognition, visual positioning and other key technologies of the visual guide AGV.1. A target tracking algorithm was proposed based on the multi-feature fusion of Sparse Representation. As the target tracking algorithm was not strong tracking robustness issues under complex conditions, the study, in the particle filter framework, introduced the sparse representation theory and the theory of affine transformation to reduce the number of the characteristics of the target state description, which could quicken up the algorithm’s operation, and had insensitivity to the target translation, rotation, scale changes and other factors. The algorithm also led up fast a particle screening method and a feature selection method based on gray and LBP characteristics. The former excluded particles with low similarity to the target, further reducing the amount of computation; the latter, through a combination of ?2 sparse representation principle, had a stronger distinction among characteristics of the target, and merged APG into the sparse equation methods, enhancing the algorithm computational speed and robustness accordingly.2. The sample learning classification mechanism was analyzed, and the study proposed a target recognition algorithm based on cascade classifier, and target tracking and recognition algorithms with Bayesian classifier. On the basis of the target tracking algorithm of structure sparse representation, Bayesian classifier classified candidate samples and determined targets’ locations, which has improved the robustness of tracking algorithm. The study further combined the tracking algorithm and the integration target recognition algorithm based on learning classification mechanism, which helped the tracking algorithm adapt the situations of target deformation and disappearance.3. An improved RANSAC estimation method was put forward. On the basis of RANSACalgorithm framework, the introduction of pre-inspection technology enhanced the operational efficiency of the fundamental matrix estimation; an evaluation function was designed to optimize the determinative mechanism inside and outside screening points, and a final preferred strategy of inside points was presented based on the density analysis on point features and the counter electrode distance, to improve the quality of end-point sets inside data. Considering the final preferred strategy of inside points, the study proposed candidate sample errors and the final M estimator for final fundamental matrix estimation, improving the accuracy of the estimation algorithm and image qualities and noise.4. Based on the intelligent vision-guided technology, an intelligent vision-based Material Handling AGV was developed by analyzing the industrial material handling needs. Through a distributed control architecture, fuzzy logic rules and intelligent guide technology, the AGV above herein could achieve automatic material handling functions.