学术文献库

With more and more machine learning methods applied to bioinspired polarized skylight navigation, the demand for polarized skylight navigation datasets is more and more urgent, which can be used to train and test machine learning methods. So, in this paper, an open polarized skylight navigation dataset is constructed for the first time. Firstly, a polarized sky model was proposed based on Sun position model, Berry sky model and Hosek sky model, which contains the information of the light intensity (LI), degree of polarization (DOP) and angle of polarization (AOP). Secondly, a polarization imaging simulation system is constructed, which can capture not only LI, DOP and AOP images, but also original black-and-white LI images in different polarization directions. Black-and-white LI images are the original data collected by actual polarization imager, so this system can completely describe the whole process of polarization imager capturing skylight polarization patterns. Above all, a polarized skylight navigation simulation (PSNS) dataset can be constructed. In addition, to facilitate researchers to build their own datasets based on their own polarized light sensors and sky models, we have disclosed the source code of polarization imager and original LI imager on GitHub.