Gaze scan-path similarity in mobile eye tracking using dynamic time warping (DTW)

Eye-tracking research, which often captures and analyses the trajectory of an observer’s gaze (commonly known as the scanpath) is crucial for understanding the dynamics of visual attention. While scanpath analysis typically relies on high-resolution data collected in controlled lab settings, it faces challenges in real-world applications, where data are often noisy. This paper explores the application of spatio-temporal scanpath similarity metrics as a measure of gaze pattern similarity, with a primary focus on dynamic time warping (DTW), alongside supplementary measures, within the context of mobile eye-tracking. We demonstrate an approach that utilises DTW via tslearn toolkit, adjusting the analysis to account for the lower spatio-temporal resolution and increased data variability inherent in mobile eye-tracking devices. Two datasets serve as the basis for our investigation: an instructed viewing set, consisting of benchmark experiment data where participants’ gaze follows predefined paths, and a free-viewing set where participants naturally engage with artwork stimuli. Our preliminary results demonstrate the robustness of DTW in capturing scanpath similarity despite data imperfections, especially when using gaze data, and to a lesser extent, with aggregated fixation data. We present a pipeline for implementing DTW in mobile eye-tracking, highlighting its effectiveness in real-world settings. While supplementary comparisons with other spatio-temporal similarity metrics are provided in the supplementary materials, the focus remains on DTW as the primary method for assessing scanpath similarity. To support further research, we provide datasets and analysis scripts publicly, alongside recommendations for applying DTW to mobile gaze data, aiming to broaden the scope of eye-tracking research.Keywords: mobile eye-tracking, temporal scanpath similarity, visual attention, dynamic time warping (DTW)