Research On Positioning Control System Of Mobile Robot Based On Fusion Of Monocular Vision And Wheeled Odometer

Positioning is the primary problem that needs to be solved in the mobile robot's autonomous navigation system.The traditional robot positioning method mainly relies on a single position sensor for positioning,such as using a wheeled odometer for aircraft estimation.However,a single sensor positioning system has various limitations in positioning effects.Now there are some solutions that combine different positioning sensors to complement each other;this paper is aimed at indoor wheeled mobile robots,from its structure and positioning.Based on the cost and effect,the positioning scheme combining wheel odometer and monocular visual odometer was selected to study,and a positioning system based on the fusion of monocular vision and wheel odometer was designed.This article has studied the following aspects:1.For monocular visual odometer,comparing image feature point method and optical flow method,using improved optical flow tracking algorithm to improve image tracking quality and tracking efficiency,by setting key frames for image information and using average parallax sliding The window strategy optimizes the image pose and position by 3 degrees of freedom,realizes the monocular vision image frame pose and solves the key frame loop detection algorithm based on the word bag model,and realizes through the global minimum loop optimization and local sliding window optimization strategy.Current frame position correction and loopback trajectory correction;2.Calculate the monocular visual odometer scale information.Converting the monocular visual odometer motion trajectory and the wheel odometer motion trajectory into the world coordinate system for time-stamp alignment,and obtaining the robot position calculated by different sensors at the same time.Multi-angle trajectory alignment is proposed.The reliability of the current calculation scale is judged by calculating the scale in various cases,and the situation that the local position error is too large is avoided,and the absolute scale of the monocular visual odometer is obtained.3.For the sliding of the wheeled mobile robot and the single visual scene,the singleeye visual odometer and the wheeled odometer pose are proposed to reduce the impact of the wheel slip and the single scene on the positioning accuracy of the robot;4.The word bag model is detected on the key frame of the image.Through the strict loopback detection process,the visual loop is obtained.The global minimum loop optimization strategy and the local sliding window optimization strategy are used to correct the current visual odometer and the fusion positioning position of the robot world coordinate system.And optimize the global loop motion trajectory to achieve the purpose of reducing the cumulative error.