| With the continuous development of hardware equipment and the emergence of big data,artificial intelligence has become an important direction for the development of mobile robots in the future,and how to apply deep learning algorithms to mobile robots has become a new difficulty.On the one hand,despite the continuous development,the high computing performance hardware equipment is still relatively large,which will seriously increase the volume and load of mobile robots.On the other hand,the accuracy of deep learning algorithms is far beyond that of traditional algorithms.However,its real-time performance is worrying.To this end,This paper studies the application of deep learning detection and tracking algorithms to mobile robots to improve the effectiveness of land mobile robots in detecting and tracking dynamic targets in near-field shallow water,and to improve the level of robot intelligence..The structure of the robot control system with calculation and motion separated is designed.In order to reduce the volume and load of mobile robots and ensure the real-time nature of deep learning algorithms,a robot hardware platform was designed using remote transmission technology.After comparing the current remote transmission technologies,this article chooses to use MJPG-stream for video transmission and Zigbee for control instruction transmission.At the same time,the robot software interface is designed based on C ++ and Python languages,which has functions such as a control protocol analysis protocol,remote video reading,data storage,and so on.A terrain information sensing strategy based on lidar is proposed,and an obstacle terrain passing strategy is designed.In order to ensure that the mobile robot moves on land,the process of detecting and tracking dynamic targets in shallow water is not disturbed by terrain,a terrain information scanning strategy based on lidar was designed.In this paper,the linkage between the steering wheel of the robot wheel and the two-dimensional lidar is used to achieve the acquisition of three-dimensional surrounding environment information.Based on the analysis of the scanning information,this article designs the judgment criteria of obstacle terrain and the passing strategies of obstacles with different distribution.An improved tracking algorithm based on detection and supervision feedback is proposed,and a visual platform search strategy is designed.Based on the analysis of the field of vision of the mobile robot’s binocular camera,this paper designs a vision platform target search strategy,and designs a steering gear rotation strategy based on different search results.In order to ensure the real-time and accuracy of the detection and tracking algorithm,YOLO-V3 and Siam-FC are selected for target detection and tracking of mobile robots after comparison.At the same time,an improved tracking algorithm based on detection and supervision feedback is designed based on the IOU(Intersection over Union)concept.The designed software and hardware platform is used to verify the design module experimentally.First,this article designed experiments to verify the effectiveness of the robot platform video and control instruction transmission.Then,in flat land and sandy land,the mobile robot scans the terrain on land and passes the experiment to verify the strategy.Secondly,the effectiveness of the shallow water dynamic target search,detection and tracking is verified by using separate experiments and multiple modules.Finally,all the designed modules are combined to verify the effectiveness of the system in this paper. |
