Design And Research Of Scene Adaptive Localization System For Mobile Robot Based On Vision-Encoder Fusion

With the rise of unmanning in recent years,mobile robots have gradually been applied on more and more fields,and mobile robots such as food delivery robots,service robots,and driverless cars have gradually entered people's daily lives.It can be seen that mobile robot technology is one of the hottest researching areas nowadays,and in the future,there will also be more and more mobile robots which can be applied in practice.And the problem of localization is the most important problem faced by mobile robots,it is also one of the most noisy parts in mobile robot technology,and it is an important research direction for robotic researchers.Only when the position of a robot is accurate,can the mobile robot carry on the subsequent work normally.This paper research on the localization technology of mobile robots which move on the flat plane.The main contents include:(1)Research on visual localization technology based on image feature points.We extracted and matched feature points of image frames which captured by camera,and then presented a mismatching points culling method based on the fundamental matrix equation and the random sample consensus algorithm,and the effectiveness of the method is verified by the next experiments.Then we present a dynamic keyframe choosing scheme,according to keyframes selected by this scheme,the position of robot are calculated by using Gauss-Newton method,then we combine random sample consensus algorithm again and enhance the accuracy of motion calculation.(2)Research on encoder localization technology of wheeled mobile robot.We analyze encoder and its localization principle and construct the arc motion model of robot.Based on the motion model and the principle of dead reckoning,we construct the reckoning and localization equation of mobile robot.After that,we verify the accuracy of encoder localization through the encoder localization experiment.(3)Research on the autonomous location algorithms of vision and encoder fusion.We construct the system state model based on encoder information and convert the result of visual localization into observation vector,and then we construct the system observation model.Then we research on the vision and encoder fusion localization through the Extend Kalman Filter,and we compare visual localization,encoder localization and fusion localization through experiments and analysis the localization accuracy of each localization mode.(4)Research on scene adaptive localization algorithm.In consideration of disadvantages of fusion localization,we improve the Extended Kalman Filter and present a scene adaptive localization system based on the Improved Extended Kalman Filter,which enables the system choosing the most suitable localization scheme for the current scene in different environment,and thus we can get a more precise localization result.And we build a simulated sensor invalidation environment and verify the accuracy and robustness of our sence adaptive localization system.