Research On Indoor Visual Navigation Technology Based On Deep Learning

With the development of Artificial Intelligence in today's society,more and more intelligent buildings are emerging.At the same time,people's demand for intelligent service robots that work indoor also is growing.At present,the navigation work of most indoor robots is designed according to a specific working environment,and there is almost no human-computer interaction during the navigation process.This article focuses on intelligent robots that perform service tasks such as indoor distribution and assistants.,aiming at the dependence of the indoor robot on the working environment and the intelligent problem in the navigation process,proposes a more general and intelligent indoor navigation strategy.First of all,based on the problem of robot navigation's dependence on the indoor environment,This paper combines deep learning technology with map construction technology,proposes an improved BiSeNet semantic segmentation neural network model,and combines the global map semantic segmentation results to build a map model.Experiments show that the improved model can not only segment and identify various target objects in multiple indoor scenes,but also achieve a better segmentation effect through the training of indoor scenes,accuracy of segmentation reached 92.2%,which is conducive to map modeling.Then aiming at the intelligent problem of robot navigation task,based on the indoor global semantic image,this paper proposes a method of constructing grid semantic map based on window operation,which integrates the semantic information into grid map,and then adds the semantic relationship to the navigation algorithm to improve the human-computer interaction of robot navigation by semantic information.Experiments show that in the navigation algorithm based on A*,using grid semantic map can improve the intelligence and human-computer interaction of robot navigation.Then,the real-time problem of navigation algorithm is discussed.This paper proposes an improved A* navigation algorithm,which optimizes the node expansion strategy of A* algorithm and improves the search efficiency by using the skip point search algorithm.Experimental results show that The improved A * algorithm has greatly improved the search time and efficiency.Finally,two different actual indoor scenes are used to simulate the overall technology.The final simulation experiment results show that the indoor navigation technology proposed in this paper can be generalized to multiple scenarios and can effectively improve the intelligence of robot navigation.After a summary analysis,the research content of this article has greater theoretical reference value and practical application value for the research on the navigation tasks of intelligent robots such as indoor distribution and assistants.