This work is the first to explore how peripheral vision, instead of central vision, can be used to read text on AR and smart glasses. We propose PeriText, a multiword reading interface using rapid serial visual presentation (RSVP). It allows users to observe the real world using central vision, while using peripheral vision to read virtual information.
Human vision can be categorized into central, para-central, and peripheral vision. The fovea provides central vision, which is the very center of gaze with an eccentricity of 2.5 (5 of the field of view), and has the highest visual acuity. Para-central vision has an eccentricity of 4, and the rest is peripheral vision, which can be up to more than 90. Peripheral vision is weaker at distinguishing detail, color, and shape, because the corresponding density of receptor and ganglion cells is lower and the representation in the visual cortex of the brain is much smaller. Therefore, unlike central vision, peripheral vision might respond differently to various text design.
To better understand the effect of different text transformation in our peripheral vision area, we conduct our first study to measure the reading efficiency under 34 different conditions, combinations of 6 text transformations, 2 retinal eccentricities, and 3 numbers of words displayed at a time for RSVP.
We investigated the performance of reading accuracy using peripheral vision with a total of 6 different text transformations: full capitalization (FC), title case (TC), serif font (SF), character-wise typewriter effect (CT), word-wise typewriter effect (WT), and control text without transformation (C). The 2 text positions are 5 and 8 retinal eccentricities. In addition, the number of words displayed at a time for RSVP paradigm is 1, 2, and 3.
In Study 1, the participants were asked to sit at a distance of 60 cm from a 22-inch LCD monitor. To ensure that text was presented at the desired eccentricities, we monitored the observers fixation using Tobii EyeX, a monitor-mounted eyetracking device calibrated beforehand. The goodput evaluated time which the period from showing the sentence to the participant reading it out was recorded, and the answers were typed into a text file immediately for further comparison. The eye-tracker started to record the eye positions for each trial at the beginning of a sentence and stopped recording right after the sentence was completed. The experiment order of the 6 text transformations was counter-balanced through a Latin Square.
The results can be reported respectively in the following design aspects:
According to the goodput results and analysis of study 1, we designed PeriText, our multiword reading interface for peripheral vision, to be 1) title case, 2) sans-serif, and with 3) animation of word-wise typewriter effect.
We then conducted our second study to compare our design and basline text presentation with the similar experimental settings in study 1.
The results showed that PeriText achieved significantly higher goodput values than control text, for both 5° (+6.7%) and 8° (+12.4%) eccentricities.
To further understand how PeriText performs in real-world settings, we designed a field study (study 3) where users wore AR glasses and walked around a university campus with 2 different levels of loadings. One is light-loading task, in which users walked on pedestrian-only walkways on the university campus and in the meantime read the text presented by PeriText. The other is heavy-loading walking task, in which users have to cross a street on campus where there would be bicycles and other vehicles. The participants were always accompanied by two experimenters all the time during the study for their safety.
The participants were asked to put on Microsoft Hololens, a head-mounted display with eye-tracker by Pupil Labs to detect the user’s conscious and unconscious glances towards the peripheral text. After user finished the 20 trials for each condition, we collected NASA-TLX index ratings, including the following 6 aspects: mental, physical, and temporal loading, as well as performance, effort, and frustration.
As for overall goodput, the reading performance of reading at 5° eccentricity is better than reading at 8° under both light-loading and heavy-loading scenario.
For both light-loading and heavy-loading tasks, the overall task load required for 5° eccentricity is lower than that for 8°.
The main limitation of our interface is that during our study, each sentence was displayed once for each trial, and the goodput value was calculated without testing if the participants comprehended the whole sentences. However, several reading studies have shown that readers do not read sequentially, but re-visit text that was previously read, which is called back regressions. In order to evaluate the actual reading speed, it is necessary to introduce such backward mechanisms into PeriText.
Also, an investigation into the relationship between PeriText and various primary tasks is interesting and worthwhile, since what users are doing with their central vision might have influence on the reading performance of using PeriText.