Intrasaccadic motion streaks jump-start gaze correction

Abstract

Rapid eye movements (saccades) incessantly shift objects across the retina. To establish object correspondence, the visual system is thought to match surface features of objects across saccades. Here, we show that an object's intrasaccadic retinal trace-a signal previously considered unavailable to visual processing-facilitates this match making. Human observers made saccades to a cued target in a circular stimulus array. Using high-speed visual projection, we swiftly rotated this array during the eyes' flight, displaying continuous intrasaccadic target motion. Observers' saccades landed between the target and a distractor, prompting secondary saccades. Independently of the availability of object features, which we controlled tightly, target motion increased the rate and reduced the latency of gaze-correcting saccades to the initial presaccadic target, in particular when the target's stimulus features incidentally gave rise to efficient motion streaks. These results suggest that intrasaccadic visual information informs the establishment of object correspondence and jump-starts gaze correction.

Fig. 1. Illustration of intrasaccadic motion streaks.

When making a saccade toward the bird on the right, its retinal projection rapidly travels from a peripheral location (fixation 1) to a foveal location (fixation 2), producing a motion streak along its retinal trajectory. This streak literally connects an object’s pre- and postsaccadic locations on the retina, possibly providing spatiotemporal continuity that may help establish object correspondence. Photo credit: Richard Schweitzer.

Fig. 2. Probing the role of postsaccadic surface features and intrasaccadic motion in gaze correction (experiment 1).

(A) Observers made a primary saccade to an exogenously cued target noise patch stimulus (one of two types). Strictly during the saccade, the target rapidly shifted positions, consistent with a 30° clockwise (CW) or counterclockwise (CCW) rotation of the entire stimulus array, so that primary saccades landed between the initially cued stimulus (the target) and the other-type stimulus (the distractor). The intrasaccadic stimulus motion was either continuous throughout 14.6 ms of rotation (i.e., 21 equidistant steps along its circular trajectory) or absent (blank screen for 14.6 ms between first and final stimulus positions). After the stimulus’s motion, pixel masks were displayed with a varying delay (surface-feature durations: 0, 25, 50, 100, 200, or 600 ms), thus occluding the identity of postsaccadic objects and limiting the observers’ ability to establish trans-saccadic correspondence using object features. (B) Stimulus motion was presented strictly during saccades, finishing, on average, 10.7 ms before saccade offset. (C) Probability of observers making a secondary saccade toward the initial presaccadic target was a function of surface-feature duration, as well as the presence of intrasaccadic motion (purple versus green points, respectively; error bars indicate ±SEM). The beige area illustrates the temporal interval in which intrasaccadic motion took place. Solid lines show predictions of the mixed-effects exponential growth model describing the increase of proportions with increasing surface-feature duration. Average parameter estimates are shown in the table below the model formula. (D) Mean differences between motion conditions for each surface-feature duration with corresponding 95% confidence intervals (CIs; gray-shaded area).

Fig. 3. Intrasaccadic motion and surface-feature duration affect the latency of gaze correction.

Secondary saccade latency across observers when making secondary saccades to either the initial presaccadic target (thick lines, circles) or the distractor (thin lines, triangles), depending on surface-feature duration and presence of intrasaccadic motion (purple versus green points and lines; error bars indicate ±SEM). The beige area indicates the temporal interval of target motion, and the vertical dashed line shows the average time of saccade offset after motion offset. Solid lines are predictions of two mixed-effects generalized additive models (GAMs) that describe the time course of observers’ secondary saccade latencies as a function of increasing surface-feature duration. Parametric coefficients of the models indicated an overall significant reduction of secondary saccade latency in the motion-present condition when saccades were directed to the target (estimate = −5.99, t = −2.21, P = 0.028) but not when they were directed to the distractor (estimate = 2.38, t = 0.49, P = 0.624). The models’ difference smooth terms further suggested a time course modulation due to intrasaccadic motion for target-bound secondary saccades [estimated degrees of freedom (edf) = 9.91, F = 2.99, P = 0.001] but again not for distractor-bound secondary saccades (edf = 1.01, F = 0.04, P = 0.836).

Fig. 4. Primary saccade landing positions influence gaze correction.

(A) d_diff is the difference between two distances, from primary saccade landing to the target and to the distractor, respectively. Positive values denote that saccades landed closer to the target than to the distractor. (B) Logistic fits modeling the relationship between d_diff and the proportion of making a secondary saccade to the target for motion-absent (purple) and motion-present (green) conditions. Panels show results for each surface-feature duration separately. Points indicate group means per 0.5-dva bin. Shaded error bars indicate 95% CIs determined by parametric bootstrapping. (C) Distributions of secondary saccade latencies for each observer. Upper and lower densities represent the motion-present and motion-absent condition, respectively. (D) Linear fits predicting inverse secondary saccade latencies to the target stimulus (transformed back to raw secondary saccade latencies) based on d_diff, surface-feature duration, and presence of intrasaccadic motion.

Fig. 5. Efficient motion streaks facilitate gaze correction.

(A) Example of a filter energy map computed by convolving the noise patch stimulus with a bank of Gabor filters. (B) The retinal trajectory of the target stimulus is the vector sum of the target’s trajectory presented on the screen and the eye position vector during presentation. We computed relative orientation by normalizing the stimulus’s orientation components using the angle of the retinal trajectory. As illustrated by motion filtering applied to the noise patch, orientations parallel to the stimulus’s motion trajectory on the retina should lead to distinctive motion streaks. (C) Results from the logistic reverse regression analysis, fitted by the multivariate GAM, averaged across all surface-feature durations. High z scores (orange) imply that filter responses in a given SF-orientation component predict the occurrence of a secondary saccade to the target when intrasaccadic motion was present (middle) and absent (left). Dotted lines demarcate the transition from negative to positive z scores estimated by the linear model corresponding to the GAM. Upper marginal means show the effect of relative orientation averaged across all SF components. The surface difference (right) clearly indicates that secondary saccades to the target (and not to the distractor) were driven by stimulus orientations parallel to the stimulus’s retinal trajectory, suggesting a role of temporal integration of fast-moving stimuli, i.e., motion streaks. (D) Results from the linear reverse regression analysis, using the inverse latency of secondary saccades made to the target as a dependent variable. High t-scores (orange) mark those SF-orientation components facilitating short saccadic reaction times. The pattern suggests that the same parallel orientations that drove secondary saccades to the target also reduced their latencies.

Fig. 6. Manipulating the congruency of intrasaccadic object motion (experiment 2).

(A) The entire stimulus array, not only the target, rotated continuously for 14.6 ms, once the onset of the primary saccade (gaze positions illustrated in blue dashed lines) was detected. Presaccadic object locations are shown as dashed circles and continuous intrasaccadic motion as arrows in the color corresponding to the polarity of the moving object. The panel thus illustrates a trial with congruent intrasaccadic rotation (in CCW direction) and congruent surface features. (B) Illustration of retinal object trajectories in a single experimental trial corresponding to the condition illustrated in (A). Eccentricities are plotted in degrees of visual angle. (C) Illustration of the five motion conditions used in experiment 2. The neutral and congruent/congruent conditions correspond to the motion-absent and motion-present conditions in experiment 1. (D) Proportions of secondary saccades made to the original presaccadic target (top) and secondary saccade latencies (bottom) averaged across both target- and distractor-bound saccades. Transparent points indicate individual means, and all error bars indicate within-subject ± SEM.