Добавил:
kiopkiopkiop18@yandex.ru t.me/Prokururor I Вовсе не секретарь, но почту проверяю Опубликованный материал нарушает ваши авторские права? Сообщите нам.
Вуз: Предмет: Файл:

Ординатура / Офтальмология / Английские материалы / Eye Movements A Window on Mind and Brain_Van Gompel_2007

.pdf
Скачиваний:
0
Добавлен:
28.03.2026
Размер:
15.82 Mб
Скачать

664

J. B. Pelz and C. Rothkopf

requiring manipulation of objects, such as 500 ms while making a pot of tea (Land, Mennie, & Rusted, 1999) and 450 ms while manipulating small parts to construct a plastic model (Pelz et al., 2000), though in both cases, the mean duration was affected by a small number of very long fixations, some lasting several seconds.

The mean extent of saccades is also task dependent. Rayner (1984) reported a mean saccade size of 1 5 during oral reading, increasing to 2 for silent reading (though measures scale with the font size). While free-viewing images on a small display, Henderson & Hollingworth (1998) reported average saccade sizes of 2 4 . Pelz et al. (2000) reported that the mean scaled with display size, up to an average of 10 for a 50” display subtending about 50 . Task-dependence extends to active tasks as well. Turano and colleagues reported mean sizes ranging from 3.1 to 5 6 for subjects walking down a hallway with the goal of locating a specific doorway (Turano, Geruschat, & Baker, 2003); Pelz & Canosa (2001) reported a mean saccade size of 11 for subjects walking down a hallway without such a specific target. Mean saccade sizes as large as 19 have been reported for complex tasks such as making a pot of tea (Land et al., 1999), though again the mean was affected by a relatively small number of very large gaze changes.

Andrews & Coppola (1999) examined the degree to which fixation duration and saccade size varied by observer and task while viewing five “visual environments”. Subjects’ eye movements were recorded in a range of conditions; in the drak, viewing repetitive textures, viewing photographs, during visual search, and while reading. They found idiosyncratic differences between individuals that covaried significantly within two groupings; 1) viewing photographs, simple patterns, and in the dark, and 2) visual search and reading. There was no significant covariance between the two groupings.

The distributions of fixation durations and the saccade magnitude represent low-level characteristics of subjects’ oculomotor behavior; monitoring the direction of gaze during a task can reveal higher-level, though subconscious, strategies adopted by subjects. Examining eye movements while walking provides an opportunity to examine how vision guides action under variable environmental and task demands. One would expect that the degree to which subjects fixate the surface on which they plan to walk will vary based on its physical characteristics, its predictability, and its visibility. In a study of how patients with retinitis pigmentosa navigate with significant visual field loss, Turano et al. (2001) had normal subjects navigate a hallway wearing a head-mounted display with an integrated eyetracker. The head-mounted display provided the same view to both eyes from a single head-mounted camera. The display’s limited field-of-view required that subjects actively acquire information that would normally be available from the periphery. Even with the reduced cues available for moving in the hallway, however, fewer than 25% of normal subjects’ fixations were directed toward the floor or boundaries between the floor and walls. Patla & Vickers (2003) studied the gaze behavior of normal subjects as they walked a predetermined path. They measured the gaze of subjects with normal vision as they walked a 10 m path under three conditions; 1) stepping on regularly spaced footprints, 2) stepping on irregularly spaced footprints, and 3) a control condition with no markings on the floor. Under all three conditions the predominant gaze behavior was what Patla and Vickers termed “traveling gaze”, in which subjects’ gaze moved at the

Ch. 31: Oculomotor Behavior in Natural and Man-Made Environments

665

same rate as, and at a fixed distance ahead of, the subjects. The travel gaze accounted for over 60% of the total gaze duration, even when there were no specific targets for foot placement. Given the predictability of the surface, and lack of explicit instructions on foot placement in the control condition, it is surprising that Patla and Vickers’ subjects focused on the floor immediately before them 60% of the time.

In the experiment presented here, we extend the investigation of task-dependent oculomotor behavior to consider the effect of environment. Incorporating the environment in the experimental design allows us to probe two issues. We can (1) examine the effect of task and environment on low-level fixation duration and saccade-amplitude metrics, and

(2) address Patla and Vicker’s (2003) surprising findings regarding gaze direction while walking without explicit directions on foot placement. By monitoring gaze as subjects walk in man-made and natural environments, we seek to understand how the varying physical characteristics and the predictability of the path affect oculomotor behavior. A paved surface should require less active monitoring because the surface is more predictable and because peripheral acuity is sufficient to detect typical variations. Walking on an uneven dirt path, however, is expected to require more active guidance, and therefore more frequent foveation.

1. Methods

Subjects performed two different tasks within two environments. The two Environments we tested were outside of an apartment complex (Man-made Environment) and in the dense woods (Wooded Environment). Within each Environment subjects were required to perform two Tasks: Free-viewing (standing in one place and looking about the scene) and

Walking. The Walking Task in the Man-made Environment was performed on a paved path, and the Walking Task in the Wooded Environment was performed on a dirt trail.

Three adult volunteers from the student population at the Rochester Institute of Technology participated and received an honorarium for their participation. Subjects had normal or corrected-to-normal vision. The experimental protocol was approved by the Institutional Review Board at the Rochester Institute of Technology. Participants provided their informed consent.

Eye-movement records were collected using a custom-built, wearable eyetracker that allowed subjects’ eye movements to be monitored without limiting head or trunk movement. The eyetracker has a scene camera placed directly above the participant’s eye. The scene camera has a field-of-view of approximately 40 × 30 . Figure 1 shows the system consisting of lightweight headgear and backpack (see Babcock & Pelz, 2004 for a detailed description of the system). The image from the scene camera was used for analysis of subjects’ gaze records, and to extract head movements (see below). Rather than perform eyetracking in real-time, the system uses a video multiplexer to record eye and scene images onto a single videotape. The video is later de-multiplexed and processed in a laboratory eyetracking system. Figure 2a shows the raw anamorphic eye and scene images; Figures 2b and 2c show the resultant scene image with overlay cursor indicating gaze

666

J. B. Pelz and C. Rothkopf

Figure 1. Wearable eyetracker records eye and scene camera images for offline analysis.

(a)

(b)

(c)

Figure 2. a) Multiplexed scene and eye images b) Wooded Environment c) Man-made Environment.

position for the Wooded and Man-made environments respectively. A semitransparent eye image is superimposed on the scene image to aid in analysis. The eye image allows blinks and track loss to be identified and be excluded from analysis.

Because gaze position was calculated off-line after completion of the experiments, the field calibration required only that the participant follow a target moved about the scene in front of the subject. The target was moved to cover a range of approximately ±20 horizontal and ±10 vertical.

The eye and scene images were demultiplexed and fed into a laboratory computer equipped with ISCAN Model PCI-726/PCI-636 processing cards. Calibration was

Ch. 31: Oculomotor Behavior in Natural and Man-Made Environments

667

completed by correlating the recorded eye and scene images collected at five positions in the field. The wearable eyetracker provides a video record of gaze superimposed over the scene image and a data stream consisting of 60 Hz horizontal and vertical eye-in-head position. The statistics of saccade amplitudes and duration and inter-saccade periods were obtained with software written to extract saccades from the gaze angle data. Because the eye movements were generated as part of ongoing natural behavior, it was rare for the eye-in-head signal to be stable due to nearly constant linear and/or rotational Vestibular- Ocular-Reflex (VOR) movements. Parsing the eye-movement patterns into “‘fixations” and “saccades” was accomplished by parallel analysis using three algorithms with different classification criteria: velocity threshold, hidden Markov model of eye velocity distributions, and an adaptive velocity algorithm (Salvucci & Goldberg, 2000; Sicuranza & Mitra, 2000). Note that in this context “‘fixation” represents periods during which a portion of the visual scene is stabilized on the retina, and not necessarily the case where the eye is stable in the orbit. Fixation duration histograms were created by counting fixations within 100 ms wide bins based on the 60-Hz eyetracker data.

Head motion was estimated from the video recording of the scene camera. A sparsemotion field was obtained by tracking specified points in successive video frames. The points were selected according to Tomasi and Kanade’s algorithm (1991) and tracked across frames using a pyramidal implementation of the basic feature tracker described by Lucas and Kanade (1981). The egomotion of the scene camera was calculated using methods described in Tian, Tomasi, & Heeger (1996). The estimated rotational head motion was aligned with the eyetracking recording via the timestamp from the video track (see Rothkopf & Pelz, 2004 for a description).

Because the subjects were free to make unconstrained head and trunk movements, a “‘fixation” in this context refers to a period during which a given point of regard is stabilized on the retina. In fact it was relatively rare for the eye to actually be still in the orbit. Only when making small eye movements during the Free-view task were traditionally defined fixations observed. The majority of the time the head and body were undergoing linear and/or rotational motion resulting in a “base” of linear and rotational vestibular eye movements upon which saccades were superimposed. For the most part, there were no objects in motion within the scene that would illicit smoothpursuit or optokinetic eye movements (although when a person did come into view s/he was invariably fixated).

Figure 3 shows horizontal and vertical eye-in-head movements for a 5-second Freeview segment in the Man-made environment. The eye moves through 60 . Except for the 300 ms fixation at 1700 ms, the eye is in constant motion. The saccades move gaze left and right as head movements and VOR return eye-in-head toward the midline. Figure 4 illustrates the horizontal and vertical head movements over the same period. Figure 5 shows the integrated horizontal and vertical gaze position over the 5-second period as the integrated eye and head movements form a series of gaze fixations and saccadic movements, despite the near-constant motion of eye and head. Comparing Figures 3 and 5 illustrates subjects’ ability to make very large gaze changes (a range of 120 is evident in Figure 5), while limiting eye-in-head movements to 60 as seen in Figure 3.

668

J. B. Pelz and C. Rothkopf

 

90

 

 

 

 

 

 

 

 

Horiz eye-in-head

degree

60

 

 

Vert eye-in-head

30

 

 

 

 

angle,

 

 

 

 

0

 

 

 

 

 

 

 

 

 

head-in-

0

1000

2000

3000

4000

–30

 

 

 

 

 

 

 

 

 

Eye

–60

 

 

 

 

 

 

 

 

 

 

–90

 

 

ms

 

 

 

 

 

 

Figure 3. Horizontal and vertical eye-in-head signals for 5-second Free-view task.

 

90

 

 

 

 

 

 

 

 

 

Horiz head

degree

60

 

 

 

Vert head

30

 

 

 

 

angle,

 

 

 

 

0

 

 

 

 

 

 

 

 

 

Head

0

1000

2000

3000

4000

–30

 

 

 

 

 

 

 

 

 

–60

–90

ms

Figure 4. Horizontal and vertical head signals.

5000

5000

In addition to extracting the low-level measures of fixation duration and saccade size, gaze behavior while Walking was classified in terms of fixation location. Each gaze period was classified as “Path” if gaze fell directly on the path within approximately 3 m in front of the subject, and as “Away” if the gaze was directed to the side or above the “Path” area. This classification was completed manually with a JVC BR-DV600U VTR controlled by a lab computer with custom software developed for the analysis. Rather than code individual fixations for gaze direction while Walking on paved or dirt trails, the video record was coded to identify periods of gaze, defined as contiguous fixations within a specified region. The performance was rated by describing the fraction of the total trial during which gaze was directed to the path so that it could be compared to the results of Turano et al., (2001) and Patla & Vickers (2003).

Ch. 31: Oculomotor Behavior in Natural and Man-Made Environments

669

Gaze angle, degree

90

60

30

0

0

–30

–60

–90

Horiz gaze

Vert gaze

1000

2000

3000

4000

5000

ms

Figure 5. Horizontal and vertical gaze signals for 5-second Free-view task.

2. Results

2.1. Fixation duration

Figure 6 shows the distribution of fixation durations for the three subjects during the Free-view task in the Man-made environment. While the mean fixation duration varied from 250 to 440 ms, the variation was due largely to differences in the frequency of fixations over 500 ms. Median fixation duration varied from 190 to 225 ms, and modal fixation durations for all three subjects were only 150 ms, as seen in Table 1.

The average fixation duration distribution across the three subjects in the Free-view task in the Man-made environment is seen in Figure 7, along with the distribution for

Relative frequency

Man-made free-view

0.4

S1

S2

S3

0.3

0.2

0.1

0.0

0

500

1000

1500

Fixation duration (ms)

Figure 6. Distributions of fixation durations for three subjects in Free-view task in the Man-made environment (Error bars represent ±1 SEM).

670 J. B. Pelz and C. Rothkopf

Table 1

Fixation duration for the Free-view task in the Manmade environment (ms)

Subject

Mean

Median

Mode

 

 

 

 

S1

250

190

150

S2

365

225

150

S3

440

210

150

 

 

 

 

Relative frequency

0.4

Man-made free-view Wooded free-view

0.3

0.2

0.1

0.0

0

500

1000

1500

Fixation duration (ms)

Figure 7. Mean fixation duration for Free-view task in Man-made (solid line) and Wooded (dashed line) environments (Error bars represent ±1 SEM).

the same task in the Wooded environment. Table 2 shows the mean, median, and modal fixation durations for the Free-view and Walking task in both Man-made and Wooded environments. Note the shift toward longer fixation durations in the Wooded environment in the Free-view task.

A two-way repeated measures ANOVA (within subjects; Environment × Task) shows a significant Environment (Man-made vs Wooded) effect in the skewness of the two

Table 2

Mean, median, and modal fixation duration for the Free-view and Walking task in the Man-made and Wooded environments (ms) across three subjects

Task

Environment

Mean

Median

Mode

 

 

 

 

 

Free-view

Man-made

375

210

150

Free-view

Wooded

440

290

250

Walking

Man-made

305

200

250

Walking

Wooded

305

210

250

 

 

 

 

 

Ch. 31: Oculomotor Behavior in Natural and Man-Made Environments

671

distributions (F1 2 = 25 44; p < 0 05), but no main effect of Task (F1 2 = 0 11, p = 0 77), or interaction (F1 2 = 0 15, p = 0 74). The primary difference between the distributions is in the shift away from very short fixations in the Free-view task in the Wooded Environment. Paired two-sample t-tests show significantly fewer fixations in the bins centered at 50 and 150 ms in the wooded than in the Man-made environments during the Free-view task (p < 0 01 and p < 0 05 respectively).

The log-fixation duration distribution approximates a Gaussian distribution, and highlights the differences in the bounded shorter fixations. As seen in Figure 8, the difference in this log-duration space appears as a narrowing of the distribution in the Wooded environment. A two-way repeated measures ANOVA (within subjects; Environment × Task) in this space revealed significant effects in the standard deviation of the log-fixation distributions. There was a significant [Environment × Task] interaction (F1 2 = 26 75; p < 0 05). The nature of the interaction is evident in Figures 9a and 9b; the significant difference in the distributions in the Man-made and Wooded environments is due to variation in the Free-view task, but not while Walking. The effect of Environment approached significance (F1 2 = 17 72, p = 0 052), and no main effect of Task was evident (F1 2 = 1 56, p = 0 34).

2.2. Saccade size

The average eye-in-head saccade size was 5 1 , varying from as low as 3 7 for free view in the Wooded environment, to as high as 5 8 for Free-view in the Man-made environment, though no significant differences were found in the distribution of saccade

Figure 8. Mean (log-scale) fixation duration in Man-made (solid line) and Wooded (dashed line) environments (Error bars represent ±1 SEM).

672

J. B. Pelz and C. Rothkopf

Relative frequency

0.50

Man-made free-view

Wooded free-view

0.25

0.00

10 100 1000

Fixation duration (ms)

(a)

Relative frequency

0.50

Man-made walking

Wooded walking

0.25

0.00

10 100 1000

Fixation duration (ms)

(b)

Figure 9. a) Mean (log-scale) fixation duration for Free-view task in Man-made and Wooded environments b) Mean (log-scale) fixation duration for Walking task in Man-made and Wooded environments (Error bars represent ±1 SEM).

sizes across Environment (F1 2 = 0 52, p = 0 49), Task (F1 2 = 0 03, p = 0 88), or any interaction (F1 2 = 0 63, p = 0 45). As is often the case with unconstrained eye movements, the mean saccade size was inflated by a relatively small number of very large gaze changes (e.g., Land et al., 1999; Pelz & Canosa, 2001); the median saccade size across all subjects and conditions never exceeded 3 .

2.3. ‘Path’ gaze distribution during walking

Figures 9b and 10, and Table 2 all indicate very little difference by (and no significant effect of) Environment on performance during the Walking task. This may be surprising, given the variation in predictability in footing between the two environments. While there was little difference in the low-level metrics of fixation duration and saccade size, Environment caused a dramatic difference in high-level gaze behavior, as measured by the fraction of gaze directed to the path immediately in front of the subjects. Recall that gaze during the Walking task was manually encoded as “Path” when subjects directed their gaze at the region where they would be taking the next few paces. Figure 11 shows the fraction of gaze coded as “Path” in the Man-made and Wooded environments. When navigating the uneven dirt path, subjects devoted 55–75% of gaze to the near “Path” region, similar to the Patla and Vickers (2003) result. One subject maintained a high fraction on the ‘Path’ region while walking on the paved walkway, but the others dropped to 15–25%, similar to that reported by Turano et al. (2001). Paired two-sample t-tests showed a significant increase in ‘Path’ fixations from 35% on the paved walkway in the Man-made environment to 62% on the dirt path in the Wooded environment (t = 2 94, p < 0 05).

Ch. 31: Oculomotor Behavior in Natural and Man-Made Environments

673

Figure 10. Distribution of saccade size for Free-view and Walking tasks in in Man-made and Wooded environments (Error bars represent ±1 SEM).

Gaze on path (%)

100

S1

S2

S3

Mean

75

50

25

0

Man-made

Wooded

Environment

Figure 11. Fraction of trial duration during which gaze is directed toward the ‘Path’ region, within approximately 3 meters of the subject in Man-made and Wooded environments. Shaded bars show fraction of gaze on ‘Path’ region for subjects S1, S2, and S3. Solid bars show the average across subjects in each environment. (Error bars represent 1 SEM).