Ground Target Search Based On UAV Multi-source Data

With the continuous occurrence of various disasters,the search and rescue for people is the most critical task faced by search and rescue personnel,which requires many human and material resources,and manual search and rescue is inefficient.The UAV is light and flexible,can realize automatic control,realize the transmission of information,and fly to areas that personnel cannot reach providing convenience for the following work.However,many problems need to be solved urgently in the actual application process.Although the drone is light and flexible,its endurance is poor;after complete the mission,the drone should be allowed to land autonomously and wait for the flight command.Excessive manual control complicates the process and squeezes search time and rescue personnel;if bad weather,the target search based on visible light conditions cannot work well.In order to solve the above problems,this article uses the DJI Mavic 2 industry dual-light version of the UAV and uses the DJI Mobile SDK to design and implement a ground target search Android application based on the multi-source data of the UAV.With the consideration of the different phone screen shapes of the current mobile phone,the application obtains the video stream after the fusion of thermal infrared and visible light and customizes the view to maximize the display of the video stream;the efficiency of the UAV power system is currently approaching the technical limit.Use drones more efficiently by planning the search area,setting the overlap rate to calculate the route interval,conducting a comprehensive search of the search area within a limited time,and reducing the number of times the drone returns to replace the battery.In the meantime,the frequency of target detection can be controlled by the sampling rate.For the target search part,the system uses the YOLOv4-Tiny network model and transplants the weights obtained from training to Android through the mobile deep learning framework NCNN and achieves a speed of 18FPS or more;For the precise landing part of the UAV,this system proposes a precise landing method for the UAV based on multiple positioning QR codes(AprilTags).By identifying the AprilTags in the apron,the UAV and the apron are obtained.At the same time,the corresponding control algorithm is designed for the situation of wind disturbance and lost view of the apron.By constantly adjusting the drone,it will land on the designated location on the apron.After completing the design and implementation,test each part and the whole of the system to verify that the system meets the requirements and design requirements.The system can significantly save human resources,reduce the probability of missing the search target,improve work efficiency,and expand the drone’s functions.