Research On Location,Navigation And Object Detection Of Service Robot Based On Deep Learning

With the prolongation of life expectancy and the decrease of the proportion of newborn population,population aging and labor shortage will become a long-standing social problem in our country,which will result in the serious problem of supporting old people that needs to be solved urgently.The development of robotics and artificial intelligence technology has made it possible to develop service robots with certain intelligence.Conducting research on service robots with convenient human-computer interaction capabilities,intelligent positioning and navigation,and object detection capabilities is of great significance for alleviating labor shortages and solving the problem of supporting old people.Focused on the aforementioned problems,the research of the positioning and navigation technology and object detection algorithm of service robots based on deep learning technology is conducted,mainly including the following aspects:(1)The motion model of the service robot is established.Firstly,the mechanical structure of the service robot is studied,and the motion model of the two-wheel differential is determined.Secondly,the coordinate system model of the service robot is established,and the positional relationship between each coordinate system is clarified.Finally,the two-wheel differential model,the odometer model,the probability observation model and the camera imaging model are constructed,which lay the foundation for the positioning and navigation research of the service robots.(2)Based on deep learning technology and the AlexNet classification network,a semantic map construction method is proposed,which can solve the problem of map construction for robots in unknown environments.Aiming at the low accuracy of indoor positioning and navigation and poor understanding of the environment of service robots,the semantic map construction method based on AlexNet classification network is studied.A dataset suitable for the application scenario is created,which includs six location categories of bedroom,corridor,dining room,kitchen,living room and office selected from the Indoor dataset.The final room recognition accuracy rate reached 98%.(3)With the deep learning technology,the lightweight object detection network of MobileNet＿YOLOX is built to meet the practical and high-precision requirements during the object detection process of service robots.Firstly,the lightweight structure of the MobileNet is used to improve the feature extraction part of the backbone of the YOLOX network.By loading the training weight of the backbone network and adjusting the network freezing strategy,the mAP of the MobileNet＿YOLOX network is improved to 77.92% on the PASCAL VOC dataset.Secondly,a network evaluation model based on the Analytic Hierarchy Process is proposed.In the network evaluation model,a criterion layer is built through evaluation indexes of detection accuracy,parameter quantity,computational complexity,and detection speed.With the criterion layer,the MobileNet＿YOLOX is compared with other advanced networks and the result shows that MobileNet＿YOLOX is a high-performance object detection network that can meet the needs of mobile terminals.Finally,the indoor furniture dataset is established,and the MobileNet＿YOLOX network is trained based on the method of transfer learning.The mAP of the MobileNet＿YOLOX network is improved to 97.38%.The results show that the improved network has good portability.(4)Based on the ROS operating system,a simulation model including a service robot and an indoor home environment are built,and a simulation experiment research is carried out,which includes key information extracting based on voice interaction,motion control based on key action information,positioning and navigation based on key location information,and object detection based on key item information.The experimental results show that the key action information extracted through voice interaction can improve the convenience of human-computer interaction of the service robot,the location and navigation based on the extracted key location information from the semantic map layer can enhance the service robot’s ability to understand the environment,learning of semantic regions can make the service robots effectively distinguish the same object in different scenes and improve the accuracy of object detection.The proposed semantic map construction method based on the AlexNet classification network can solve the problem of low information acquisition of traditional occupancy grid maps and improve the environment understanding ability of service robots.The designed MobileNet＿YOLOX lightweight object detection network can realize a small number of parameters,fast detection speed,high recognition accuracy and other performance improvements.The research is of great significance to promote the development of service robots.