Research On Bionic Positioning And Orientation Based On The Polarization Image

In this paper,taking the autonomous navigation task of unmanned platform without satellites as the application background,drawing on the biological navigation mechanism and polarized light sensitive mechanism,this paper focuses on the polarization HDR image preprocessing method for bionic navigation,the environmental feature extraction method based on the polarization image,the biomimetic orientation method for polarized HDR images and the biomimetic localization method based on empirical knowledge of navigation.The main research work and results of the paper are summarized as follows:Firstly,aiming at the imaging requirements of bionic visual navigation under complex lighting conditions,an image enhancement method based on bionic polarization image sensor is studied.The theory proves that the bionic polarized image sensor has better HDR perception characteristics than traditional cameras.Combined with the principle of polarization vision measurement and camera imaging mechanism,a HDR imaging model based on the bionic polarization image sensor is deduced,and a single-frame polarization HDR construction method is proposed based on the model.The experimental results of imaging in HDR environment show that,compared with the single-frame HDR construction method of visible light images,the proposed single-frame polarization HDR construction method has a great improvement in Peak Signal-to-noise Ratio,Structural Similarity and other metrics.To improve the image quality in the environment with lower degree of polarization,a "model + learning" polarization image enhancement method is proposed,which realizes the fusion enhancement of the single-frame polarization HDR reconstruction model and the inverse tone mapping network.The proposed method can effectively improve the whole image quality.Secondly,aiming at the problem of feature extraction under complex lighting conditions,the research on environmental feature extraction based on bionic polarization image sensor is carried out.Experimental results have verified that under complex lighting conditions,compared with visible light images,the single-frame polarized HDR image constructed based on the bionic polarized image sensor can improve the repeatability rate of various visual geometric feature detectors.The influence of different combinations of polarization information on the extraction of environmental semantic features is studied.The experimental results show that the fusion of polarization information under complex lighting conditions has better precision,recall and other metric performance than single visible light information for environmental semantic features(especially vehicles).Thirdly,aimed at the problem of polarized light orientation in the sky under complex lighting conditions,a biomimetic orientation method based on polarized HDR images is proposed.The limited dynamic range of polarization measurement under the condition of illumination change is analyzed,and a polarization information measurement method based on multi-exposure fusion is proposed,which effectively improves the adaptability and accuracy of the algorithm compared with the traditional polarization orientation method.Aiming at the nonlinear response of the pixels of the bionic polarization image sensor to the light intensity of the scene,a self-calibration method of radiation value based on exposure ratio and exposure time is studied,which improves the measurement of polarization information when the light intensity difference is large due to random factors such as occlusion and the cloud.In order to further improve the adaptability of bionic polarized light orientation,combined with the sky region extraction algorithm based on the consistency of atmospheric degree of polarization distribution and morphological filtering,an adaptive exposure adjustment algorithm based on local information entropy and heuristic segmentation is proposed.Indoor light source conditions and orientation experiments in outdoor natural environments demonstrate that the proposed method can effectively overcome illumination changes and improve the adaptability of the bionic polarized light orientation.Finally,aiming at the problem of autonomous positioning of unmanned platforms without satellites,a navigation semantic map with empirical knowledge is constructed.A bionic positioning method based on navigation experience and knowledge is proposed.It can effectively suppresses the accumulation of errors.Aiming at the problem of the difficulty in correlating the onboard image of the unmanned platform with the navigation semantic map data,a method is proposed to describe the environment where the feature points are located by comprehensively using the heading information,the semantic information,the navigation attribute information of the feature points and the mapping relationship between each information.The semantic encoding function of the semantic topology structure can effectively reduce the number of singular values in the feature matching process,thereby improving the positioning accuracy of the unmanned platform and the adaptability of the navigation algorithm.The heteroscedasticity problem of multi-source heterogeneous information in the visual/inertial integrated navigation system based on the navigation semantic map is studied.A weighted combination optimization model is designed for outdoor global positioning of unmanned platforms.The air unmanned platform experiment proves that the final positioning error is reduced from 64.12 m to 15.29 m compared with the visual inertial system.