Bio-Inspired Polarization Compass for Solar Azimuth Prediction Under Clear and Cloudy Sky Conditions

Sunlight becomes partially polarized due to Rayleigh scattering while passing through the atmosphere. Many insects such as ants and beetles utilize the polarization information of skylight for navigation. Unlike magnetic compass and GPS, this scheme is free from interference and unlike inertial schemes, its error does not accumulate with time. In recent years this navigation scheme has received a lot of attention for navigation of aerial and terrestrial vehicles. Development of a polarization compass that can provide accurate heading information in different weather conditions will benefit a wide range of applications. Here we report the application of linear regression for implementation of a polarization compass. The model is trained with real sky images and then used to predict the solar azimuth in both clear and cloudy sky conditions. The root mean square errors for both conditions were less than 1°. We also compared the performance of the proposed scheme with that of Hough transform and support vector machine, which have been successfully utilized for the same application in the past. Linear regression outperformed Hough transform for all sky conditions considered, and its performance was comparable with support vector machine. However, unlike the other two methods considered, the accuracy of linear regression will increase significantly when trained with a large set of sky images. Therefore, with sufficient training, linear regression can be a promising option for implementation of polarization compass.