Research On Bionic Slam Algorithm Based On Visual Odometer

With the rapid development of science and technology,more and more mobile robots appear in all walks of life.Mobile robots are used in actual production.The first problem to be solved is the autonomous positioning and map construction of mobile robots.At present,the algorithms for this problem are mainly based on probabilistic algorithms,among which the Most Expected Algorithms,Particle Filter Algorithm and Kalman Filter Algorithm and their optimized related algorithms are common.However,with the rapid development of disciplines such as neuroscience and anatomy,more and more research in the scientific field has begun to obtain design principles and innovative ideas through the biological world from the perspective of bionics.At this time,a problem was also raised in the field of mobile robot SLAM "How to make mobile robots more intelligently solve their own positioning and map construction problems like mammals?".Studies have found that most mammals obtain visual scene information through the eyes,and then transfer the visual scene information to the visual area in the brain,process it through the visual system,and then pass it to the hippocampus and other areas in the brain to complete their own positioning and map construction through the navigation system in the brain.From the perspective of bionics,this paper imitates the overall idea of mammals to complete their own positioning and map construction,draws on the research results of neuronal cells in the hippocampus in the field of neuroscience to complete the establishment of a cell model,and uses the visual milestone in the classic visual SLAM system To establish a vision system,in order to complete the research on the problem of synchronous positioning and map construction of mobile robots from the perspective of bionics.The main research work of this article is as follows:This article first analyzes the role of head cells,grid cells and place cells in the mouse brain navigation system and its own working mechanism to establish the corresponding cell model.A single head direction cell can complete the characterization of a single direction.In this paper,multiple head direction cells are used to complete the characterization of the spatial direction information to establish a head direction cell model.Although the grid cell attractor model is simple,it lacks physiological basis.Instead,it uses the self-motion information of the mouse as an input.The grid cell oscillation interference model established by superimposing multiple oscillation waves in the olfactory cortex of the hippocampus is more convincing.Using the research results of location cell information from grid cells,a grid cell to location cell model is established,and based on this model,a visual landmark model is used to construct an environmental representation map model.Nodes are generated when the mobile robot is running,and the environment is characterized by the node set.At the same time,a grid cell reset mechanism is set during the map construction process to reduce the number of nodes and increase the accuracy of the map.Then optimize the visual odometer.Due to the high environmental requirements of the camera itself,in the process of acquiring depth information of the visual scene,the acquired depth information has a lot of environmental noise and outliers,which will negatively affect the position estimation.Good bilateral filtering to deal with noise and outliers.In the actual environment,the images collected by the camera have problems such as less texture information and too concentrated distribution of texture information.On the one hand,the image information entropy is used to filter out images with too little texture.On the other hand,the feature points are in the form of a quad tree The division makes the distribution of feature points extracted from the image more uniform,and then use the exact matching pairs selected by the coarse screening as the sample set,and obtains an accurate matching model through the RANSAC algorithm to match the feature points of the two images.It is more accurate and improves the accuracy of pose estimation.Finally,the optimized visual odometer is used to replace the dorsal visual pathway in the visual system to complete the motion perception function and provide self-motion information.Use image feature matching to replace the ventral visual pathway to complete the object recognition function,build a visual system model,and combine the visual system model and the navigation system model to establish a bionic SLAM model based on the visual odometer.In this model,the use of the ventral visual pathway in the visual system to improve the object recognition function improves the grid cell reset mechanism,further increasing the accuracy of the map.In order to verify the algorithm model proposed in this paper,Kinect cameras,notebook computers and mobile robots are used to build a hardware experimental platform for experimental verification.Collect data through the camera and send it to the host computer for processing.The experimental results show that the bionic SLAM model based on the visual odometer constructed by combining the visual system and the navigation system can achieve the characterization of the environment,with higher accuracy and stronger robustness.