Posture Analysis Of Moving Objects Based On Multi-view Information

In the past ten years,with the development of aerospace,autonomous driving,military weapons and other fields,it is very important to analyze the moving objects in the multiple fields of view.Posture parameters can not only reflect the motion state of moving objects,but also have extremely important practical significance for analyzing various physical parameters of moving objects,experimental determination and troubleshooting.In this paper,the current research status of posture analysis is introduced in detail.Taking the aircraft as the research object,the key point localization algorithm based on cascade convolutional network is designed,and the posture angle regression algorithm based on BP neural network is designed.And the experimental object verification of the moving object aircraft is carried out in multi-view environment.The main research contents and innovations of this paper are summarized as follows:1)Aiming at the posture problem of moving objects,combined with the geometric characteristics of moving obj ects,the coordinate information of moving objects is taken as the input feature and the posture angle information is used as the algorithm model of the output label.The training precision and speed of the attitude analysis model are improved.2)Taking the aircraft as an example,based on the idea of cascade convolutional network,an algorithm for locating the key points of the target object on the two-dimensional image is designed,and the distance metric is introduced to improve the traditional loss function.At the same time,the posture angle regression algorithm is designed based on BP neural network,and the evaluation model is evaluated by using the average absolute error and fitting coefficient.The simulation results show that the average absolute error of the attitude angle predicted by the depth learning based attitude analysis algorithm can be controlled between 0.60°and 1.6°,the absolute error of the predicted attitude angle and the true attitude angle of about 98.9 1%of the samples can be controlled within 3°,which has higher prediction accuracy than the traditional attitude analysis algorithm.3)A multi-field experimental device was built to evaluate the prediction accuracy of the attitude analysis algorithm designed in this paper in multi-view field environment.The pre-processed plurality of video sequences are input into the trained algorithm model to predict the attitude angle of the target object.The experimental results show that in the multi-field environment,the attitude angle of the moving object can be predicted more accurately.