Research On Robot Path Planning Based On Image Restoration Technology

In the research of mobile agent,it is one of the key problems that the agent make path planning in the range of detecting road.The path planning of intelligent robot make possible for itself to reach destination accurately,quickly and safely.The SLAM of the path planning reflect the ability of the agent to interact with the surrounding environment in the process of motion.However,the intelligent machine will be interfered by external factors in the process of path planning,which will lead the image that collected by visual sensor to blur,affect the identification of obstacles,and eventually the failure of path planning.In order to solve the above problems,firstly,the image restoration technology is used to Denoise the image interfered by external factors.Secondly,the environment model is constructed,and what the intelligent algorithm of convolution neural network is used to identify the path obstacles,which effectively avoids the defects of low fault tolerance and insufficient redundancy of the traditional algorithm.Finally,the reinforcement learning decision network is used for path planning.In this paper,the image restoration technology is applied to the agent,which is combined with the neural network intelligent algorithm,and the path planning problem of the agent is studied under the laboratory simulation experiment.Firstly,the images(atmospheric turbulence,wind-sand dust,heavy fog weather)collected by visual sensors are eliminated blur and noise by blind convolution method in this paper.Purpose of doing this to improve the previous noise processing technology,and changes from the method of convolution between the external noise signal and the original image pixel to the processing method of selective amplification of the external noise environment.The image can be restored to the original noise-free and clear state as much as possible.Secondly,the input form of the processed image is transferred to the column vector,then it is input into the convolution neural network of deep learning.In addition,the structure of the traditional convolution neural network is changed slightly in this paper,which makes the pooling layer more prominent and reduces the operation burden of the convolution neural network.The data trained in this paper come from the self-built data set,which can accurately grasp the characteristics of the obstacle body,and use convolutionneural network to identify and mark the objects(obstacles)in the image.Finally,the marked obstacle background image is used as the input,and the obstacle coordinates in the image are extracted and stored as database files by using MATLAB,and the shrinking obstacle is reconstructed to the new Cartesian coordinate system to mark the starting point and the target point at the same time.Continue to analyze whether the coordinate points of obstacles,target points and starting points of the new map have Markov property,and then whether the Markov process can be constructed.finally,Markov analysis is carried out,the agent is defined and Markov matrix is constructed.Using Python to write reinforcement learning DQN algorithm program for agent path planning.Finally,the moving route and distance are recorded,and it is analyzed that the feasibility of reinforcement learning algorithm in path planning.The final experimental results show that image restoration technology plays an important role in agent path planning and acquisition system,and the clarity of image decides that the follow-up work of agent path planning can be carried out normally.Secondly,it is verified that the combination of image restoration technology and artificial intelligence algorithm in path planning is feasible and the system with reinforcement learning control decision algorithm can run stably for a long time in the complex and changeable big data environment.