Research On Indoor Object Recognition Based On Vision

In order to improve the recognition accuracy of mobile robots for indoor target objects,this paper studies the map construction,navigation and target object recognition and positioning functions of mobile robots for indoor environment.Firstly,the mobile robot completes map construction and path planning,and then uses visual sensors to identify indoor objects.Summarize the current mobile robot for indoor object recognition difficulties,and give a reasonable solution.In order to improve the recognition accuracy of Goog Le Net algorithm,the network model structure of Goog Le Net algorithm is improved.The improved Goog Le Net algorithm and Goog Le Net algorithm are compared and analyzed with 6 different object images.To solve the problem of low accuracy of SSD algorithm for target object recognition,the network model structure and the size and quantity of default box corresponding to feature graph in SSD algorithm are improved.The improved SSD algorithm,original SSD algorithm and related target detection algorithm were compared and analyzed using PASCAL VOC 2007 test data set and images downloaded from network.On the basis of ROS Navigation function package,the Navigation system of mobile robot is designed,and 2d lidar is used to realize map construction and Navigation of mobile robot in indoor environment.The improved Goog Le Net algorithm combined with visual sensor is used to realize indoor object recognition.Based on the improved SSD algorithm and the vision sensor,the indoor object can be recognized and located and the actual size of the object can be estimated.In the indoor environment,the mobile robot completes the map construction and navigation,and uses the improved Goog Le Net algorithm to realize the indoor object recognition.