| With the rapid development of UAV technology,UAV applications in various fields are becoming more and more extensive.In the design and development of UAVs,simulation platforms play a crucial role in simulating UAV behavior and performance in real-world environments for systematic verification and testing.However,when checking the simulation platform,it is necessary to ensure that the simulation model accurately represents the developer’s concept description and specifications to ensure the credibility and validity of the simulation results.However,the drone may encounter obstructions in actual operation,such as buildings,trees or other aircraft,which will have an important impact on the perception and navigation ability of the drone.Therefore,accurate identification of UAVs has become a key task in the verification process of UAV simulation platform.Therefore,the main research content of this paper is as follows.(1)We propose a dual-mode fusion motion target recognition algorithm for the case of non-occlusion.Background differencing and template matching algorithms are widely used methods for motion target detection and recognition.However,using a single algorithm often fails to achieve the desired results.Therefore,this paper presents a dual-mode fusion motion target recognition algorithm that can fully utilize the advantages of both algorithms.First,the video is subjected to background initialization and a background model is established.Then,the foreground information of the moving target is extracted using background differencing technique,followed by template matching to accurately locate the motion target.Additionally,the Kalman filter is employed to further enhance the tracking performance of the unmanned aerial vehicle(UAV)target.Finally,comprehensive testing and evaluation of the proposed algorithm are conducted.The experimental results demonstrate that the proposed algorithm performs well and maintains high recognition accuracy even in situations involving changes in the shape of the target.It exhibits superior detection accuracy and is capable of effectively handling variations in target shape.(2)A trajectory prediction network model is proposed for the case of occlusion of unmanned aerial vehicles(UAVs).Traditional prediction models often rely on linear equations or Gaussian process models,which are limited in handling nonlinear and complex prediction problems.In this paper,a model based on generative adversarial networks(GANs)is introduced to predict the trajectories of UAVs.The LSTM network is utilized to encode the sequential information of the trajectory data,extracting trajectory patterns.After pooling,the decoded trajectory is generated by the decoder,producing a synthetic trajectory sequence for the UAV.The prediction of future UAV trajectories is achieved by adversarial training between real and fake trajectories.The training results demonstrate that the generator and discriminator are continuously optimized and engaged in a game as the iterations progress.Furthermore,experimental data confirms that the proposed network model exhibits favorable trajectory prediction performance and capability.This research provides strong technical support for the validation and application of UAV simulation platforms. |
