Design And Implementation Of Drone Patrol System Combining Image Stitching And Object Detection

With the rapid development of technology and the continuous cost reduction,drones have become a part of our lives and been widely used in various fields,such as remotesensing mapping,agricultural monitoring,security patrols,and military reconnaissance.Compared with traditional security methods like human patrol or camera monitoring,drone patrol has many advantages,including flexibility and large monitoring range.However,there still exist some problems such as the limited range of vision and the absence of timely reports on abnormal situations.Considering the shortcomings of the current drone patrol systems,we have implemented a drone patrol system combining the image stitching algorithm and the object detection algorithm.The system includes a mobile client and a server.The mobile client is oriented to the drone operator.By connecting the drone remote control,the mobile client can obtain the real-time video stream returned by the drone,periodically sample patrol images from the video stream and upload them to the server.The server includes front-end and back-end.Oriented to managers,the front-end of the server can display the status of patrol tasks in real time and will automatically alarm when abnormal objects are detected,thereby helping managers to make timely response measures.At the same time,the server front-end provides data labeling function,with which managers can strengthen the object detection model by manually labeling data to optimize the patrol effect.The server back-end is used to process requests sent by the client and the server front-end.It can store patrol images and return information about patrol tasks.The image stitching algorithm and object detection algorithm are used to process the patrol images while eliminating motion ghosts during the process of image stitching to optimize stitching effect.The thesis introduces the drone patrol system in the aspects of requirements,design,implementation and testing.Firstly,the thesis analyzes the drone patrol system from the perspective of functional and non-functional requirements,and introduces the system use cases in detail in the functional requirements analysis.Secondly,the thesis introduces the system architecture diagram of the drone patrol system,and divides the system into four functional modules in terms of their respective functions,including patrol images management module,patrol images processing module,patrol statuses management module and detection models management module.The function modules introduce the main functions of the module through flowcharts or timing charts,and then introduce modification or optimization to the core algorithms of the system.Next,the thesis shows the specific implementation of each function module through code snippet examples,and gives the actual operating interface of the system.Finally,the thesis uses the DJI Mavic 2 PRO drone to test the functionality and performance of the proposed drone patrol system,and has achieved desirable results.