Research On Attitude Estimation Of Methods Based On Multi-scale And Adaptive Gradient Descent

In recent years,the standards of people’s living have been continuously improved,and the use of cars has become more frequent.But when we use cars,many problems will still be encountered.Therefore,the emergence of smart cars has also become the focus of research by experts,scholars and industry professionals.Researchers use advanced science and technology to optimize the control of vehicles to ensure that vehicles drive more safely.The perception of the vehicle’s own motion state is the focus of intelligent research,and the estimation of motion parameters is its key.Among them,the indicators describing the vehicle’s motion state mainly include the displacement,speed,and attitude of the vehicle.Therefore,it can be seen that the attitude of the vehicle is o ne of the important parameters describing the movement of the vehicle.In this paper,the attitude of the intelligent vehicle parameters will be selected for research.For attitude estimation,an attitude estimation method based on multi-scale and adaptive gradient descent is proposed,which is carried out from three aspects: dual attitude model,multi-scale attitude estimation method and adaptive gradient descent attitude estimation.The specific research methods and innovations are :Because quaternion attitude estimation has global non-singularity,the calculation is relatively large.Although the calculation amount of high-order Rodriguez parameter attitude estimation is relatively small,singularity will appear in the calculation process.For the advantages and disadvantages of the two methods,an estimation model based on dual attitudes is designed,which can be switched to different attitude models under different states of the vehicle’s posture resolution.An attitude estimation method based on multiple scales is proposed.The principle of wavelet multi-scale is introduced.On the basis of the HOMRP model,combined with the Kalman filtering method,a multi-scale algorithm is then used to optimize this model.And a simulation experiment is used to verify that the method is effective.Using this method to calculate can obtain a more accurate attitude angle.An adaptive gradient descent attitude estimation method is proposed.The principle of gradient descent is introduced,and based on the quaternion model,the gradient descent method is used to optimize this model.At the same time,the PLS algorithm is used to dynamically determine the step size of the gradient descent to achieve adaptive effect.In addition,mean filter is also used to deal with the data input of the accelerometer to achieve the purpose of suppressing motion noise.Then the simulation experiment of this method is conducted,which proves that this method can make the calculation of attitude angle more accurate.Finally,in this thesis,IMU inertial navigation sensors and computers were used to carry out in-vehicle tests.Comparing adaptive gradient descent with commonly used quaternion methods,we can see the effectiveness of adaptive gradient methods for attitude calculation.By comparing multi-scale method with HOMRP method,it proves that the multi-scale method can improve the accuracy of the attitude solution.Finally,the three methods are comprehensively compared,proving that the method proposed in this paper can improve the accuracy and convergence of the attitude estimation of the sports car.If using this method in a smart car,we can control vehicle movements better and ensure the safety of smart cars.