The Research And Implementation Of Navigation Technology Of Indoor Mobile Robot Based On Monocular Vision

Navigation is an indispensable technology for achieving autonomous robot movement and has been an important direction in the field of robotics research.Navigation based on GPS system achieves better results,it is not suitable for mobile robot navigation in narrow environments such as indoor and alleyways due to occlusion and large positioning errors.Simultaneous localization and mapping(SLAM)is a method that can solve the problem of mobile robot navigation in indoor environments.However,the current conventional Kalman filter-based SLAM algorithms still cannot meet the practical needs.This study builds a brain-like navigation system,which mimics animals' navigation mechanism because animals have amazing navigation ability.The core of the study is an improved RatSLAM algorithm and a cognitive map-based navigation method.RatSLAM uses the head-facing cells,location cells,and raster cells encoding mechanisms in the brain navigation mechanism to mimic the navigation ability of animals.However,the method uses a simple template matching approach to achieve scene recognition and complete closed-loop detection.For scenes with single environmental information,it is prone to generate scenes mismatching and mapping failure.In this thesis,the ORB algorithm is used to extract features from environmental images,generate descriptors,and use the Hamming distance between descriptors to measure the similarity between images to improve the accuracy of image matching.Then the descriptors are clustered into words to build a tree dictionary,which converts the sequential query image retrieval in the original algorithm into a tree dictionary query and improves the speed of image retrieval.Experimental results in public datasets and real scenarios show that the loopback detection scheme proposed in this thesis improves the accuracy and recall of image matching of the RatSLAM algorithm,while the proposed method does not increase the retrieval time as the robot's exploration time in the environment grows,ensuring that realtime processing can always be achieved.Meanwhile,this thesis builds a monocular visionbased autonomous navigation system for mobile robots based on the improved RatSLAM to realize mobile robot navigation based on brain-like cognitive maps.From the constructed cognitive map,the Dijkstra algorithm is used to achieve global path planning,find the optimal global path from the current position to the target position,and complete local path planning using the dynamic window approach to achieve the mobile robot to reach the target position along the global optimal path avoiding dynamic obstacles.