Research On Mobile Robot Bionic On SLAM Algorithm Based On Growth Self-Organizing Map

The problem of autonomous navigation of mobile robots in unknown environments is actually the problem of mobile robots positioning themselves and describing unknown environments,Researches on these two issues have produced Simultaneous Localization and Mapping(SLAM)techniques.Most SLAM research is based on mathematical probability methods,commonly used laser navigation,inertial navigation,infrared navigation and WLAN navigation,etc,but due to the cost of sensors and the problems of the algorithm itself,these navigation methods cannot be widely used.Visual SLAM(Visual SLAM,VSLAM)is a camera-based visual SLAM positioning method,which uses the acquired image information to describe the features to estimate the robot's posture and then complete the construction of the environment map.Compared with the traditional SLAM method,the VSLAM method has better real-time performance and higher positioning accuracy,and has become a research hotspot in the SLAM direction in recent years.This article takes the VP-SLAM visual navigation method that simulates mouse navigation as the research direction,and uses the pose cell network and visual scene cells to complete the traditional SLAM positioning and composition tasks,and aiming at the problems that VP-SLAM is susceptible to environmental light changes and low positioning accuracy,an improved feature matching algorithm VP-SLAM model is proposed to reduce the impact of environmental light changes;A growth self-organizing neural network VP-SLAM model based on improved matching algorithm is proposed to improve the positioning accuracy of the model.Through in-depth study of the VP-SLAM model and model experiments for the laboratory environment,it can be seen that the matching of scene images in the VP-SLAM model is completed by the sum of absolute differences(SAD)algorithm,the SAD algorithm is more sensitive to changes in light.In the case of direct light and obstruction by obstacles,the VP-SLAM model will cause mismatches in the scene,the laboratory experiment results of the VP-SLAM model show that when there are too many mismatches,the positioning error of the robot becomes larger,and the loopback detection effect is not ideal,resulting in the generated experience map not matching the actual trajectory.In this paper,a SURF feature matching algorithm is used to replace the original model of the SAD algorithm for visual image matching.First,the feature descriptor generated by the SURF algorithm is used to complete the feature point extraction of the image,put the extracted feature vector into the visual template library to assist the VP-SLAM model to complete the matching between the current template and the historical template in the visual template library,and then complete the correction of the characterization error of the pose cell network activity,effectively improve the accuracy of template matching,and finally get excellent experience map.Secondly,it uses a growth self-organizing map(GSOM)algorithm,and can use its unsupervised learning model to quickly build environmental topology maps.Then,according to the activation response characteristics of location cells,a response model of location cells is constructed.Based on the GSOM and response model,by collecting environmental data,the response response of the location cell activation and neurons in the output layer of the neural network is established,the topological map of the environment is constructed through the GSOM neural network,and then the distance information of the road sign The activation response of the position cell is realized to estimate the position of the object,so that the positioning accuracy is further improved,and finally a more accurate experience map is generated.In this paper,the VP-SLAM model of the improved algorithm is experimentally verified through the built hardware platform.The experimental results show that:the introduction of SURF feature matching algorithm effectively improves the shortcomings of the original model's sensitivity to light changes,significantly increases the accuracy of template matching,and obtains a more accurate experience map;The proposed GSOM algorithm effectively constructs the environment topology map,and uses the position cell response model to estimate the position,which reduces the positioning error of the VP-SLAM model and improves the ability of the VP-SLAM model to locate and compose the indoor environment.