Research On Indoorscene Recognition Method For Mobile Robot Based On Two-dimensional Distance Measurement

As an important derivative product of artificial intelligence,intelligent mobile robot has greatly changed the way of human production and life.Scene recognition is an important branch of intelligent mobile robot research.The realization of many functions of mobile robot,such as positioning,navigation and path planning,is based on the good scene recognition ability.The research of mobile robot indoor scene recognition based on the two-dimensional ranging information is a hot spot.Lidar has the advantages of high resolution and wide scanning range.In this paper,a method of scene classification based on laser ranging information and local receptive fields based extreme learning machine is studied for the situation of various indoor scene mixing.Sonar sensors have the advantages of low cost and easy installation.For the situation in which the indoor arrangement is constantly changing due to human behavior,the paper studies a method of indoor scene recognition based on sonar ranging and feature fusion.The main work of this paper is as follows:Firstly,the development of intelligent mobile robot and the changes it has brought to human society development are investigated.And the background and significance of indoor scene recognition technology are summarized.The main research methods and difficulties of indoor scene recognition technology are deeply analyzed.Then,under the Linux,the 3D simulation environments of kitchen,toilet,living room and bedroom are built by using the robot operating system and the Gazebo simulator to simulate the real life scene.At the same time,the Pioneer 3-AT mobile robot model is built and the lidar ranging system and sonar sensor ranging system are arranged.In simulation environments,a large number of virtual laser radar ranging information and sonar ranging information are collected to generate the data set.Considering the shortcomings of discrete and narrow features,the ranging data is transformed into binarized images,which can effectively represent the size and shape characteristics of the indoor environment.The traditional classifier SVM has the disadvantages of complex training,long time consuming and easy to produce local optimal solution.In this paper,the local receptive fields based extreme learning machine is used as the classifier instead,which avoids the iterative adjustment of parameters and improves the efficiency greatly.Finally,the scene recognition technology is studied by using virtual lidar ranging information.And the experiments are carried out by using the real laser ranging data set DR Dataset to verify the validity of the method and the reliability of the simulation environment.In addition,the method to classify the indoor scene which is constantly changing is studied based on virtual sonar sensor ranging information.And four adjacent distance measurement data are fused,which improves the classification accuracy effectively.