Research And Implementation Of Aircraft Pose Estimation Algorithm

Object pose estimation is an important task in computer vision.The pose information of the aircraft can be obtained through pose estimation.This information can provide a certain basis for predicting the aircraft’s next move.At present,most of the pose estimation algorithms for aircraft are based on manual feature extraction or deep learning technology.The latter related algorithms can greatly reduce labor costs and are more effective and robust.However,these algorithms generally have problems such as too slow speed,large prediction error or poor adaptation of mobile platforms.In order to improve these problems,this thesis optimizes the existing algorithm to achieve a pose estimation algorithm for aircraft that have better performance.This thesis also designs and implements an airborne pose estimation system for aircraft.The main work of this thesis is as follows:（1）In view of the problem that the current aerial pose dataset of aircraft is not ideal,this thesis implements a well-established dataset.This dataset provides a solid foundation for model training and testing,helping the algorithm achieve a good result.（2）The pose estimation network for aircraft has been built and optimized.This thesis optimize the residual block in the network and adds the attention mechanism.The optimized model reduces the amount of parameters by 19% under the condition that the pose estimation effect is improved.（3）Using Py Qt5 and the pose estimation algorithm for aircraft to design and implement an airborne pose estimation system for aircraft.The system that has a clear interface and simple operation can perform accurate pose estimation on the input image data.（4）Deploy and adapt the system on the NVIDIA mobile computing platform Xavier.Finally,the system can run stably at a processing speed of 21 frames per second.