Human Pose Recognition And Pilot Posture Tracking

Pose recognition is one of the topics in computer science and language technology,and its purpose is to interpret human pose information through models and algorithms.Human body pose recognition technology is an indispensable part in the field of computer vision due to its broad application prospects in the fields of personal health care,environmental awareness,human-computer interaction and monitoring systems.At the same time,in the field of aerospace,the operation and performance of the pilot in the cockpit are of great significance to the entire voyage.Many aspects of the driver’s condition may affect the correct operation.Therefore,this article will focus on the attitude estimation problem in detail,and study the application of driver attitude tracking in weightless environment.First,this paper proposes a method of deep sparse Gaussian process to solve the pose estimation problem of the human body.This method solves the limitations of the standard Gaussian process,constructs a multilayer network structure by recursion,and introduces pseudo input and output as auxiliary variables,reduces the computational complexity by sparse learning,and then introduces the hidden variable model into the network structure.The derivation between layers through the variational method makes the generalization and learning ability of the model have been greatly improved compared to before,and the better flexibility allows it to solve more complex problems and better Is applied to the problem of attitude estimation.For the attitude tracking problem of the driver in the space capsule,this article mainly uses deep learning to track.Four different network models are pre-researched: Mask R-CNN,Deeper Cut,RMPE,and Open Pose.These model frameworks and important innovations are analyzed in depth,and they are applied to the driver pose recognition scenario.The structure of the model and the characteristics of the algorithm are used to analyze and compare the application of different methods.At the same time,according to the real-time requirements of driver detection during driving,by improving the Open Pose model,a Fast Open Pose model was proposed to improve the original network structure and control the number of phases in the structure,and the structure of the convolution kernel of 7 × 7 has been replaced to reduce the amount of calculation while maintaining the range of receptive fields,and then a residual network was added to suppress the hidden danger of gradient disappearance in the deep network caused by the convolution kernel,and the way of extracting the network by sharing parameters and replacing features reduced the model calculation amount.Therefore,the detection efficiency is greatly improved while ensuring the detection performance of the model,and the real-time monitoring requirement of the driver’s posture is guaranteed.Finally,the operation status of the pilot in the aircraft cabin is tracked through the application of different models.In this paper,the Mask R-CNN model is used to track the driver’s hand(also its key operating part),and the convolution operation is used to simultaneously implement the target detection and instance segmentation tasks of the driver’s hand.In addition,the adjusted Quick Open Pose model is used to track the driver’s arm state,and the entire operation process is monitored from three-dimensional space through the calculation of advance data.As a result,comprehensive tracking of pilot operations in the aircraft cabin is achieved.