Salient features in gaze-aligned recordings of human visual input during free exploration of natural environments

During free exploration, humans adjust their gaze by combining body, head, and eye movements. Laboratory experiments on the stimulus features driving gaze, however, typically focus on eye-in-head movements, use potentially biased stimuli, and restrict the field of view. Our novel wearable eye-tracking system (EyeSeeCam) overcomes these limitations. We recorded gaze- and head-centered videos of the visual input of observers freely exploring real-world environments (4 indoor, 8 outdoor), yielding approximately 10 h of data. Global power spectra reveal little difference between head- and gaze-centered recordings. Local stimulus features exhibit spatial biases in head-centered coordinates, which are environment-dependent, but consistent across observers. Eye-in-head movements center these biases in gaze-centered coordinates, leading to elevated "salient" features at center of gaze. This shows that central biases in image feature distributions in "natural" photographs are not a property of environments, but of stimuli already gaze-centered by the photographer. Further central biases in laboratory subjects' fixation distributions do not result from re-centering of the eyes but are an artifact of display restrictions. Hence, our findings demonstrate that the concept of feature "saliency" transfers from the laboratory to free exploration, but also highlight the importance of experiments with freely moving eyes, head, and body.