Gaze tracker manual

Each saccadic event is displayed in the y-t plane. Finally, the tangential force is displayed 4. In addition, the power of this type of display becomes readily apparent. We are able to determine not only what the subject was looking at and for how long, but also what force they were applying over the duration of a fixation.

By examining how the applied forces change with time and the shifts in gaze that accompany them, we can get a qualitative feel for the gaze-motor gaze motor characteristics of the subject. Figure 4 shows hows data taken from the experienced subject. However, after approximately 30 seconds the frequency of fixations increases, and this is followed by an increase in applied forces in the tangential direction. We compare the duration of fixations, the variance of fixation positions, the characteristics of the forces applied and identify a general correlation between the eye movement and force.

As noted earlier, we shall refer to the experienced subject as Subject 1 and the intermediate as Subject 2. Furthermore, for the purpose of direction consistency, we will call the x direction of the gaze as tangential and the y direction as axial.

A two sample t-test and a chi-squared test were performed on the fixation data. This would make sense as we may guess that the fixation durations are an inherent property of the individual subjects. As a result, we pooled the data for each subject across all three trials.

The pooled distributions as illustrated in Figure 6 are highly skewed. A comparison between the two subjects illustrates that Subject 1 in general has longer fixations with a wider distribution of fixation duration. Subject 2 on the other hand has much shorter durations indicating a more frequent shift in gaze. We cannot speak to the importance of any point of regard for the particular subject per se.

However, the longer the duration, the fewer the number of fixations that occur in a given trial. Also, a longer fixation implies more time that the subject has to attend to the information from that location in space. So we may imply a greater importance to points of regard that occur at long intervals. We can invert fixation duration in order to get a distribution of fixation frequency as shown in Figure 7. This is not a true spectral distribution--one could argue that this is a somewhat artificial construction as the fixations defined in analysis are determined as quick shifts in gaze rather than smooth movements.

However, we use this method of viewing the data to compare against the spectral distribution of the tangential forces as we shall see shortly. There are two main reasons for choosing the tangential direction only. First, the task was performed primarily in the tangential direction. Second, there was little variation in the axial direction that could be accounted for as "eye- in-head" movement.

As shown in Figure 9, each subject had a tendency to move their view axially as the task progressed thereby mitigating the importance of shifts in that direction--we couldn't tell them apart from re-alignments of the head. Figure 6: Comparison of skewed fixation distributions between subjects. Data from all trials for each subject is pooled. Red lines indicate the median of the distribution.

Whiskers extend out to the 90th percentiles. Outliers beyond the 90th percentile labeled with a red cross. We calculate the variations as follows. First, the mean for each trial is calculated.

The variation is defined as the square root of the squared difference of the gaze position from the mean for each fixation. Thus it is not the variance of a distribution, distribution, but the absolute value of the distance of each fixational point of regard from the mean of each particular trial.

These data are presented as pixels in the image frame horizontal by vertical. These data are binned and the distribution of eac each subject for each trial is plotted in Figure We can see that there is a general trend of lower variation in the more experienced subject Subject 1 than the intermediate inte subject Subject 2.

Figure 7: Normalized Histograms of fixational frequency for each subject. Figure 8: Distributions of the tangential variations for all the trials. The whiskers extend to the 90th percentile of the distribution. Outliers are represented as red crosses. The tangential force is also an indication of the grinding power as the greater the magnitude, the more energy involved in the material removal Odum et al.

Finally, there is variability in the distributions of the tangential forces between subjects that were large enough to analyze statistically. Therefore, we chose to analyze forces in the tangential rather than the normal and axial directions.

Figure 9: Plot of all fixation points for all trials. Positions are reported in pixels on the original by pixel field of view. The origin is located in the upper left hand corner of the image frame. Notice the shift in the means of trials can be due, for instance, to the relative position a subject stood with respect to the grinding sample.

Figure a Plot of normal and tangential forces from a typical grinding experiment. The height of the bars represent the mean force for each trial with the standard deviations indicated. Figure 11 shows the power spectral density of the tangential forces for the two subjects. These densities are averages across the three trials for each subject.

We can see that not only did Subject 2 apply a higher static force, as shown in Figure 10 b , but these forces contain energy dispersed over a broader bandwidth. This is especially true for frequencies above 1. The modes in these spectra are located at: Mode 1: 0. These modes arise through the proprioceptive interaction of the human subjects with the natural dynamics of the mechanical system.

We cannot say for certain which regime may predominate in this particular frequency band. However, the clamped test article is extremely stiff, and the grinding wheel was rotating at rpm. Mechanical resonances are most likely absent at such frequencies. It would not be imprudent to say that the force response in the 0 to 5Hz bandwidth is dominated by the characteristics of the gaze-motor system.

Figure Power spectra of tangential forces averaged over all trials for each subject. Data are plotted between.

The modes of interest are labeled 1 through 4. Our experimental setup cannot measure where in space a force was applied, only its components along the principle axes of the triaxial load cell. However, we can track how the tangential force changed, and correspondingly, how the eye movements change in the same direction.

A close inspection of Figure 5 reveals an interesting quality in the gaze-motor behavior of Subject 2. We can see that the fixations proceed along with, and in relative phase with, changes in the tangential force.

The eyes are moving along with the hand as it presses down into the sample. As a means of comparing the frequency characteristics of the gaze and tangential force properties of each subject, we have overlaid the frequency distributions of the fixations and power spectral density onto a single plot. These plots are shown in Figure We can see from Figure 12 b that Subject 2 has a peak in the force spectrum at Mode 2, which overlaps with a similar peak in the fixation distribution.

Mode 2 corresponds to the back and forth sweeping motion of the hand and is the primary motion that is visible during the performance of the task.

There are also, however, overlapping peaks at Mode 3, implying a correspondence between gaze and force at that frequency as well. Similarly, Subject 1 also displays an overlap of peaks at Mode 1. In contrast, there are a number of peaks in the fixation distribution that do not seem to correspond to any modes of the tangential force. The difference between these two figures exemplifies the contrast in performance between the two subjects.

Subject 1 applied less force, with fewer changes in tangential fixation than Subject 2. Likewise, Subject 1 was less likely to change fixations beyond the initial sweeping hand modes.

The other fixational modes may have importance outside of the gaze- motor feedback loop, or may be indicative of another process not captured in these data. More importantly, the information gleened from the GMSTC could be used to impute an importance to particular saccades in the scanpath, and indicate what eye movements are likely to correspond to significant hand movements.

Figure Frequency distributions of the fixations and power spectral densities of tangential forces of a the experienced subject and b the intermediate subject. As we have shown, the characteristics of each subject vary.

However, both the experienced and intermediate subjects show a correlation of eye movement to force in some degree indicating an importance between the two. This leads for the design of a more thorough experiment that would examine the eye movements for a wider range of tasks possibly two dimensional grinding and how the forces change between "important" saccades. The possible rule for quantifying the importance of a saccade has to be a function of the experience of the practitioner.

Since the skills involved in complex manual tasks need the close integration of multiple processes, the difficulty for the data analysis then lies on how to deal with data collected from processes of different nature. The paper shows our attempt to tackle such a difficulty. By extending existing methods, such as scanpath methods and space-time cube method, we develop a visualization method with which we are able to illustrate gaze and motor data in a unified representation.

Such a representation enables us to carry out meaningful comparisons of data collected from humans with different skill levels. We are currently working on collecting data from a larger and more comprehensive sample population which may also capture confounding effects on behavior such as gender or age. We are also continuing to develop an algorithm to learn a joint gaze-motor model, and study the implications of the learned model to the design of human-robot systems.

References Bauer, A. Human—robot collaboration: a survey. Humanoid Robot. Begum, M. Visual attention for robotic cognition: a survey.

IEEE Trans. On 3, 92— Borji, A. State-of-the-art in visual attention modeling. Pattern Anal. On 35, — Modeling top-down visual attention in complex interactive environments. Man Cybern. On 44, — Cristino, F. ScanMatch: A novel method for comparing fixation sequences. Methods 42, — Duchowski, A. Eye tracking methodology: Theory and practice. ACM, pp. Erez, T. Goldberg, J. Goldstein, E. Cognitive psychology: Connecting mind, research and everyday experience.

To run this software, I recommend you install a gaze tracker, such as Tobii Eye Tracker, and a head tracker such as webcam. It also requires Windows 7 or higher. For head tracking my favorite option is Enable Viacam eViacam. It tracks your head position using a standard webcam. If you prefer infrared tracking, then TrackIR is really good. It tracks an infrared headset you clip to a hat or your headphones.

It works best with contacts instead of glasses because the light from the Tobii Eye Tracker can reflect off your glasses and distract it. Head tracking is included in version 4C of higher.

The 4C is also compatible with more computers since it supports USB 2 and 3. You also need a way to click, and there are several choices. A regular keyboard hot key can be chosen to act as a mouse click. Alternatively, you can use a USB switch. I like the Specs switch from Ablenet. Ablenet offers a wide variety of switches for people with different needs. Other options include the Fragpedal to click with your foot, Dragon to click by speaking, or you can use dwell clicking with eViacam if you are unable to use any of these devices.

Install the latest release using our installer. Calibrate each and run them in the background. Lastly, start the Precision Gaze Mouse program. Set your X and Y axis speeds so the mouse moves across a few inches of screen when you comfortably rotate your head.

Head tracking is used to point at small targets within this radius. As a result, you get natural mouse control with both speed and precision. Control your mouse by looking at your screen Download. We combine the speed of eye gaze tracking with head tracking for single-pixel precision. Learn more about how it works in our user manual.