Target Detection And Positioning Method Based On UAV

In recent years,with the advancement of drone technology,its application fields have become increasingly broad.Combining drone remote sensing with target detection and positioning can make greater contributions to fields such as security,search and rescue,and measurement.Due to the performance limitations of small drones,they have difficulty satisfying the high real-time requirements and good detection effects of the algorithm.At the same time,target detection and positioning require suitable viewing angles.Therefore,the engineering aspect should design intelligent drone flight platforms to meet these requirements;in terms of algorithms,research should focus on target detection and positioning algorithms for drone remote sensing images.Both complement each other,having theoretical value and practical significance.Previous research shows that it is difficult to guarantee the positioning accuracy of drones in complex scenes.To improve the accuracy of target positioning methods,it needs to be achieved under the premise of ensuring computational speed.Introducing high-performance depth estimation methods can effectively reduce positioning errors caused by complex terrains.This paper studies target detection and depth estimation of remote sensing images based on deep learning methods,respectively.To improve the accuracy of target positioning,theoretical analysis,method research,and experimental verification work are carried out.The main research contents are as follows:1)Drone Intelligent Flight Platform: Target detection and positioning tasks rely on drone flight control.This study constructs a multi-drone collaborative operation platform,analyzes video images in real-time,and provides effective feedback on flight status to improve detection and positioning accuracy.The multi-drone platform is closer to real-world application scenarios,solves practical problems and has practical value.2)Depth Estimation Method: In complex environments,traditional monoculardepth estimation has difficulty finding accurate and stable reference baselines.Converting predicted relative depth into absolute depth requires setting reference planes or scales,typically using the ground as a baseline.However,complex environments often have multiple or non-planar surfaces,making it difficult to find accurate reference planes.This paper studies monocular-depth estimation and designs a progressive depth estimation method,using drone flight information as a basis,and designs a mapping relationship from relative depth to absolute depth.Experiments show that this method can effectively complete depth estimation tasks in complex scenes.3)Target Positioning Method: Monocular cameras lack depth information in images,necessitating the design of positioning methods suitable for drones equipped with monocular cameras.By establishing a geometric model between the camera and the drone,a target positioning method based on monocular vision is proposed.This method can calibrate the camera’s intrinsic and extrinsic parameters,obtain the target’s position information within the image,and further convert it into a position in the drone coordinate system.Experimental verification reveals that this method can effectively improve the positioning accuracy and correctness of the drone,especially in complex environments where its performance is even better.