Research On Bioinspired Polarized Skylight Attitude Determination

Since the discovery that some insects can detect polarized skylight for navigation through the compound eyes,bioinspired polarized skylight navigation has become a new research hotspot in navigation.Although some achievements have been made in bioinspired polarized skylight navigation,there are some problems.This paper refines the relevant problems,and focuses on polarized skylight attitude determination.The main contents include:(1)In view of the lack of analytical skylight polarization model considering the influence of meteorological factors and the distribution of light intensity(LI)in the sky,an improved analytical skylight polarization model is constructed based on Berry sky model and Hosek sky model(describing the LI information in the sky).This improved model considers the influence of neutral points,atmospheric turbidity,ground albedo,light wavelength,and can describe not only the distribution of degree of polarization(DOP)and angle of polarization(AOP)in the sky,but also the distribution of LI.In addition,a polarization imaging simulation system is designed,which can obtain not only DOP,AOP and total LI images,but also the original LI images in different polarization directions.Finally,based on the proposed improved skylight polarization model and polarization imaging simulation system,a polarized skylight attitude determination simulation system is constructed.A large number of simulation experiments have been carried out using this system,and a public polarized skylight image simulation dataset is established for attitude determination.(2)Aiming at the difficulty of the current polarized skylight orientation determination method to effectively calculate orientation under severe weather conditions,the idea of polarized skylight orientation determination using support vector machines(SVM)is proposed,which transforms the orientation determination problem into a binary classification problem and considers the influence of noise and cloud using soft-margin SVM to determine orientation.The polarized skylight orientation determination experimental platform is designed and constructed.In different weather conditions,such as sunny,cloudy and overcast,the experiments have been carried out,and the anti-interference ability of SVM method is highlighted.Moreover,the influence of carrier tilts(polarized light sensor does not point to the sky zenith)on bio-inspired polarization orientation determination is studied,and this type of error is corrected.(3)Aiming at the difficulty of the classical polarized light attitude determination methods to reflect the process of insect nervous system encoding and processing polarization information,referring to the unique polarization information processing neural system of insects,artificial neural network(ANN)models for polarized skylight attitude determination are designed.Firstly,a polarized skylight orientation determination ANN is proposed.This neural network has specific dilated convolution,which directly extract DOP and AOP from the original LI information in different polarization directions.Besides,the exponential function encoding of orientation is designed as the network output,which can better reflect the insects’ encoding of polarization information,and improve the accuracy of orientation determination.Secondly,the network is further improved,and a solar position determination ANN is proposed,which can determine not only the solar azimuth angle but also the solar altitude angle.Finally,the effectiveness of the networks is verified by training and testing on the constructed polarized skylight image dataset.(4)In view of the shortcomings of some theoretical models and methods in the current polarized skylight three-dimensional(3D)attitude determination,it is theoretically proved that Rayleigh sky model contains only single solar vector information,which contains only two independent scalar pieces of attitude information,so it is impossible to determine 3D attitude simultaneously in real-time based only on Rayleigh sky model.We discussed the shortcoming of the current real-time polarized skylight 3D attitude determination methods based only on Rayleigh sky model.Furthermore,a polarized skylight 3D attitude determination algorithm based on template matching and social spider optimization is designed.Then,it is verified that the proposed improved skylight polarization model contains3 D attitude information under common meteorological conditions,but 3D attitude information is very difficult to determine.So,a polarized skylight/accelerometer double vector fusion 3D attitude determination model is designed,and 3D attitude determination experiments have been carried out.The experimental results show that the root mean square error(RMSE)of three attitude angles does not exceed 0.21 degrees.