VSS 2004 - Vision Science Society 2004 Conference Abstracts
Keane, B. & Pylyshyn, Z. (2004) Tracking behind occluders is not based on predicting likely reappearance locations. Vision Sciences 2004, Sarasota, FL.
Purpose. In Multiple Object Tracking (MOT), observers follow a specified subset of identical visual objects that move independently about a display. Using a variant of MOT in which all objects instantaneously disappeared and reappeared during tracking, it was found (a) subjects track better when objects reappeared at the locus of disappearance than when they reappeared in their extrapolated trajectory, and (b) tracking deteriorated as a function of the distance that objects displaced from that locus, provided that they did not displace in traveled trajectories (Keane&Pylyshyn, 2003). However, reappearance-locations might be predicted only if local occlusion/disocclusion cues enhance the appearance that movement continues while objects are invisible. The present study examines whether our previous findings can be replicated when objects move behind and emerge from (virtual) occluding edges. Method. The standard MOT task was modified so that part way through each trial, all objects on the screen disappear as though going behind occluding edges. Either 450ms or 650ms later, objects gradually reappear either (a) as if they had stopped moving during total occlusion ("minimal-move" condition); or (b) as if they had moved during total occlusion ("predicted-move" condition). Results. Previous results using instantaneously disappearing/reappearing objects were replicated. Subjects tracked better in the minimal-move than the predicted-move condition for both disappearance durations. In the predicted-move condition subjects tracked better in the 450ms duration. There was no significant difference between performance at the different durations for the minimal-move condition. Conclusion: Observers do not appear to track objects predictively under any conditions examined so far; they appear to keep track of where objects disappear but cannot extrapolate to where objects are likely to reappear.
Credit: This research was supported by NIMH Grant R01-MH 60924 to ZP
Bullot, N. J., Droulez, J., Morvan, C., Pylyshyn, Z. W. (2004) Keeping track of objects while exploring a spatial layout with partial cues: Location-based and direction-based strategies. Vision Sciences 2004, Sarasota, FL.
Last year we reported (Bullot, Droulez & Pylyshyn 2003, VSS abstracts) results of an experiment (using a Modified Traveling Salesman Problem or MTSP) in which subjects were able to visit once and only once up to 10 invisible targets, which were known only from directional segments pointing toward each token target. We hypothesized that two distinct strategies were used to keep track of the targets: A "location-based strategy", in which subjects kept track of where targets were located, and a "segment-based strategy" in which subjects focused on the segments themselves and kept track of which ones corresponded to targets that had been visited. We now report new MSTP experiments which further explore these two strategies. Subjects observed a computer-controlled vehicle that visited a number of targets. As before, the location of the targets was invisible; only the direction of each the target from the vehicle was displayed. Two forms of display were studied: "allocentric", in which the vehicle moved on the screen and the targets remained fixed in screen coordinates, and "egocentric," in which the vehicle remained stationary at the center of the screen while the targets moved (as if the environment was being viewed from a frame of references fixed at the vehicle). At the end of each trial, the directional segments were extended to the edge of the screen and subjects were asked to perform two tasks by referring to these pointing segments. In the "status task" observers had to indicate for each segment whether the corresponding target was visited or not. In the "pointing task" they had to locate each target along its directional segment using a mouse. The first task is measures subjects' use of what we called the "segment-based strategy" while the second measures their use of the "location-based strategy". Results showed that the "status task" can be performed very reliably with 4 or 6 targets in both viewing condition, whereas the accuracy of the pointing task remains low, especially in the egocentric condition. These results are consistent with our hypothesis according to which performance in the MTSP task (especially with larger numbers of targets) can only be reached by a segment-tracking strategy, which can be viewed as a deictic strategy (Ballard et al., 1997; Pylyshyn, 2001).