Research On 3D Scene Reconstruction Technology Of Mobile Robots For Medical Environment

High-resolution three-dimensional scene reconstruction is the basis of mobile robot environment perception and analysis.At present,mobile robot applications oriented to medical environment have achieved certain research results.However,due to the dense display of medical environment instruments,equipment and beds,the accuracy of mobile robot three-dimensional reconstruction and path planning is directly affected.Neural Radiation Fields(NeRF)encode three-dimensional geometry and color information of objects to build a high-fidelity three-dimensional reconstruction view and store it as the weight of neural networks in lightweight models,which is ideal for memory-constrained systems such as mobile robots.Therefore,based on NeRF,this paper conducts an in-depth study on three-dimensional reconstruction and path planning of mobile robots in medical indoor environment.The main work contents are as follows:(1)Aiming at the shortage of medical indoor environment dataset,a camera observation model based on motion recovery and reconstruction was constructed,indoor scene image data of a university hospital in Guizhou was collected,data preprocessing process was designed,and medical environment datasets such as observation ward corridor,ward 1 and treatment room of the school infirmary were established,with a total number of 952 images.(2)In view of the dense display of objects in the medical indoor scene,an improved multi-layer perceptron based neural radiance fields 3D reconstruction method(IP-NeRF)was proposed.The MLP-based multi-feature joint learning method and improved gating mechanism were used to optimize the two feature extraction network structures of NeRF,which solved the problems of ambiguity and alialiation in the reconstructed view.The algorithm was verified by using four evaluation indexes on three public datasets.The experimental results show that the comprehensive performance of IP-NeRF is significantly better than other six mainstream algorithms.Three-dimensional scene reconstruction was carried out on the self-built medical environment data set,which was used as the basis for subsequent grid map generation,providing a solution for mobile robots to carry out threedimensional reconstruction in the indoor environment with dense object display.(3)Nerfstudio framework was used to limit the boundaries of the medical reconstructed scene to generate a 3D grid map with better performance and less memory consumption,in view of the complex and varied characteristics of light travel in the medical scene.On the basis of NeRF-Navigation,greedy information gain strategy and dynamic weight strategy were introduced,and A dynamic weighted A* path planning method with implicit environment adaptation was proposed.Four evaluation indexes were used to verify the algorithm on the grid map of medical scene.The experimental results show that the improved method can effectively reduce the curvature of path planning.The linear motion and obstacle avoidance performance of the mobile robot in the medical indoor scene were raised.(4)Taking the observation ward corridor,ward 1 and archives room of the school infirmary as the research objects,the above theoretical research was integrated based on Py Qt,and a mobile robot 3D planning system for medical environment was developed,which realized the high quality 3D mapping and path planning,and laid the foundation for its practical application in the medical environment.