Research And Application Of Visual SLAM Based On "end-edge" Deploymen

Visual SLAM is a key technology in Augmented Reality,and mobile devices equipped with monocular cameras are an ideal application platform for Augmented Reality applications.However,visual SLAM is a computationally intensive task,which is limited by the computing and storage capabilities of mobile devices.Local computing of visual SLAM tasks on mobile devices is difficult to meet real-time requirements.In recent years,with the introduction of the "end-to-edge" collaborative computing model,mobile devices as "ends" can utilize the computing resources of edge nodes as "edges" to improve the computing efficiency of mobile devices.In response to the difficulty in meeting the real-time operational requirements of visual SLAM for mobile device performance,this article has done the following work:(1)A visual SLAM method based on "end-to-edge" collaborative computing is proposed to solve the problem that mobile devices computing capabilities cannot calculate SLAM tasks in real time.This method decomposes visual SLAM tasks,and mobile devices perform ORB features and matching,and select key frames based on the amount of inter frame motion to send to edge nodes.Edge nodes construct local maps while performing loop back detection to optimize camera posture and map in real-time,achieving "end-to-edge" collaborative computing,providing real-time and accurate posture estimation services for mobile devices.(2)Design and implement a visual SLAM system based on "end-to-edge" deployment.The system uses Kube Edge edge computing framework,which is composed of mobile end and edge end.Mobile devices capture real scene images and select key frames from them.Edge nodes calculate accurate camera pose tracking,build local maps,and return the pose tracking results to the mobile device.Mobile devices draw camera pose tracks based on the pose tracking results to achieve real-time pose tracking in real scenes.This article designs and implements a visual SLAM system based on "end-to-edge" deployment.The test results show that the algorithm design can provide accurate and real-time pose estimation services for mobile devices,and the system can stably track indoor and outdoor scenes.Therefore,the visual SLAM system based on "end-to-edge" deployment has certain theoretical significance and application value.