Research On Vision Localization Algorithm Aided By IMU Sensor

With the continuous development of computer equipment and the increasing demand for indoor positioning,visual localization technology is favored by indoor positioning services due to its light equipment,high cost performance,wide range of applications,and rich information.Among them,visual SLAM(Simulations Localization and Mapping)has attracted much attention from scientific research scholars.Visual SLAM overcomes the signal limitation of GPS in indoor positioning.In the absence of prior environmental information,it establishes an environmental model during the movement and estimates its own movement at the same time.However,monocular SLAM has the problems of indeterminate scale and greatly affected by factors such as ambient light and texture.Therefore,a fast ACE(Automatic Color Enhancement)visual positioning algorithm assisted by IMU(Inertial Measurement Unit)is proposed to solve the problem of unfixed monocular SLAM scale and excessive environmental impact.The main research is as follows:1.Propose an improved fast ACE algorithm visual odometer based on the enhancement of the image pyramid idea.This algorithm improves the visual positioning that is negatively affected by poor illumination and weak texture environment.The ORB-VO and LK-VO visual odometer front-ends were built,and the original image,different enhancement algorithms,and the performance of the algorithm in this paper in the visual odometer were compared through experiments based on three sets of images.It is found that the algorithm in this paper can effectively improve the robustness of visual positioning,the number of extracted feature points has increased by multiples,and the matching tracking has increased by 7% to 25%.2.The fast ACE monocular vision and IMU fusion localization algorithm based on feature points is used to solve the problem of low vision positioning accuracy caused by the monocular vision due to weak environmental texture and tracking loss and uncertain scale.The front end of the algorithm uses the Gauss Newton method to solve the vision and IMU translation constraints to restore the initial scale,the back end uses a sliding window mechanism to construct a tightly coupled residual model to improve the accuracy of monocular vision.Compared with the ORB-SLAM2 algorithm based on the living room scene with uneven texture distribution,it solves the problem that the weak texture environment is easy to be lost;Compared with ORB-SLAM2 and VINS-mono,the algorithm in this paper is based on the public data set Eo Roc,which solves the problem of the overall error of ORB-SLAM2 due to the large scale problem,and the positioning accuracy is slightly improved compared with VINS-mono.3.Based on Ubuntu16.04 to install ROS software platform.Use MATLAB and OPENCV to calibrate the camera internal parameters and distortion coefficients.Set up the visual odometer and SLAM positioning system involved in the above experiment.Experiments show that the proposed algorithm can effectively restore the monocular vision scale,improve the accuracy and robustness of vision positioning,and is of great significance to the study of monocular vision and IMU fusion positioning algorithms.