Research And Implementation Of AGV Positioning Technology Based On Machine Vision

With the rapid application of computers and artificial intelligence in industrial automation,AGV(automated guided vehicle)is widely used in various operations in industry as a highly integrated transportation equipment.Real-time positioning is the premise of the normal operation of the AGV system,and it is also the core technology of the research and development of AGVs.Vision-based positioning technology can use relatively simple equipment and positioning solutions to complete real-time positioning of mobile robots such as AGVs,which has high accuracy,good adaptability and low cost.Real-time environmental perception,global image acquisition and preprocessing,image matching and positioning are the core technologies of the positioning system,and are also the main factors that determine the system response speed,positioning accuracy and robustness.Therefore,this paper mainly focuses on in-depth research on AGV positioning technology based on machine vision.First,the Zhang Zhengyou calibration method was used to determine the internal parameters of the camera to reduce the impact of camera distortion on the positioning system,and the laser rangefinder and monocular camera were used as environment sensing devices to complete the real-time environment sensing and storage of the entire AGV operating route.The scene image in front of the AGV and the distance from the front wall are collected every 1 meter on the AGV running route,to complete the construction of the global environment map database.Content-based image features and local feature point data are extracted to generate a small and simple structured pre-processing database for fast AGV positioning.Secondly,direction gradient histogram and improved ORB(Oriented FAST and Rotate BRIEF)method are used to complete global image feature and local feature point extraction.The histogram of direction gradient mainly reflects the edge information of the image,and can better describe the appearance and shape of the image.It is used as a research method for global feature extraction in the research plan.The feature point extraction of the ORB algorithm was improved;by using the SURF(Speeded Up Robust Features)algorithm to extract image feature points,construct a four-layer image pyramid and calculate the determinant value of the 26-point Hessian matrix of the three-dimensional neighborhood of the feature points to obtain scale-invariant features.After the rough matching is completed,the GMS(Grid-based Motion Statistics)algorithm is used for feature point registration to eliminate the mismatched points.The improved ORB method improves the processing speed,and is more robust than the ORB and SURF methods in image scale,curl change,brightness change,and blur processing.Finally,a set of AGV real-time positioning system based on machine vision is designed.Feature matching is mainly realized by two-level retrieval system,which are content-based image retrieval and precise retrieval based on improved ORB and HOG features respectively.Precise matching was conducted to realize AGV positioning.The actual test shows that the method proposed in this paper is characterized by a faster response speed,lower hardware requirements and stronger adaptability while satisfying the positioning accuracy as compared with classic SLAM method.