Fig. 2: Description of the computational model.
From: Celestial compass sensor mimics the insect eye for navigation under cloudy and occluded skies
a Adding the pulses from two photodiodes (PDs) with orthogonal polarised light filters can approximate the light intensity (I); subtracting them approximates the polarisation Fresnel ratio (PFR). Normalising the PFR with light intensity provides a good approximation of how well the polarity of light is aligned with the orientation of the polarisation axis analyser (PAA, p). The difference between I and p results in their celestial integration (c), which indicates how well the PAA is aligned with the brightest celestial body in the sky. b An example of the intensity and polarisation measurements of all the PAAs in a clear sky condition. Each measurement can be multiplied with a unit vector directed towards the azimuth of the PAA (see red and purple arrows), which can also flip by negative measurements (e.g., for p). c The simple addition of all the vectors of each unit type results in a vector that has the direction of the solar azimuth (anti-solar for p). d The measured responses of the different types of units (solid lines) in two different sky conditions (clear and partially cloudy with occlusions from trees) and the estimations of our model using only I (red), p (purple), or c (blue). Small triangles denote the median estimation across a set of 360 homogeneously distributed facing directions (one per degree) of the sensor; circular lines denote the first and third quartiles of the distribution; the green line highlights the solar azimuth.
