Bionic Navigation Algorithms Based On Polarization Vision

To solve the problem of autonomous navigation in satellite signal rejection conditions,this thesis study on the bionic navigation methods based on polarization vision.The main research contents are carried out in the bionic sensors and bio-inspired navigation algorithms,by learning form the mechanisms of biological navigation.The major contributions and innovation points are as follows:(1)The optimal orientation algorithm based on polarization vision is proposed.First,the least-squares estimation algorithm is derived for skylight polarization measurement,based on the principle of polarized light detection.Second,the skylight polarization pattern based on Rayleigh scattering model is established,and the sun vector is established as an optimization problem of finding the minimum eigenvector.This method improves the orientation accuracy by comprehensively utilizing all of the effective area of the polarization pattern.Last,a method based on random sample consensus(RANSAC)is proposed,to solve the problem of orientation under foliage environment with heavy occlusions.It shows that the polarized light compass can provide accurate heading angle by seeing only a tiny part of the skylight(minimum of 0.028%).(2)The optimum design and integration method of the polarization sensors are explored.First,the optimal alignment of the polarizers is analyzed,which provides a theoretical reference for sensor design.It shows that the optimal configuration is to space the analyzers out evenly between 0° and 180°.Second,the error model of the polarization sensor with four synchronized cameras is established,and the joint calibration algorithm is introduced to improve the image registration accuracy.Last,the design and integration method of a pixelated polarization camera is explored,and the calibration method is also carried out based on the sensor's error model.The allowable range of installation error is analyzed,which laid the foundation for the follow-up research work.(3)A new method for image enhancement based on polarization vision is explored.First,when observing scenes in distance,the polarization information is used to decouple the airlight from the object radiance,which makes the recovered scenes with better contrast.It also shows that polarization information is helpful for scene identification and object detection.Second,the transmission and reflection scenes in the semi-reflective scenario are reconstructed by polarization analysis,which suppresses the mutual interferences and makes the texture of the target clearer.(4)The bio-inspired methods for topological node recognition and scene feature expression are proposed.First,the node recognition algorithm based on grid cells model is studied,and the inertial/visual odometry is integrated with the SeqSLAM algorithm for node recognition and navigation error correction.Second,the feature expression method is explored based on the visual mechanism of human brain.The convolution neural network and the deep learning algorithm are used to generate the global feature of the images from different cameras.Last,the method of topological node construction is studied,and it realizes effective organization and utilization of navigation experiences.(5)The bionic navigation algorithm with heading/position constraint is proposed.The inertial/visual odometry is used for dead reckoning in three-dimensional space.The accumulated navigation errors are corrected with the heading constraint of polarized light compass and the position constraint of topological node.The bionic navigation mechanism of "Dead reckoning + Heading constraint + Position constraint" is established,and the corresponding algorithm is derived.The observability of the system under different constraints is analyzed,which provides references for related theory and applied research.The experimental results show that the proposed algorithm can significantly improve the positioning and orientation accuracy of the integrated navigation system.