A growing number of theoretical and computational models of vision assert that humans construct an internal representation of an external three-dimensional (3-D) object from a relatively small number of two-dimensional (2-D) characteristic or prototypic views. In Poggio and Edelman’s words, “having enough 2-D views of an object is equivalent to having its 3-D structure specified” (p. 263). One source of evidence that 3-D objects are represented in memory by multiple characteristic views is provided by investigations of the way objects are visually inspected. This research has consistently shown that when people are asked to learn objects for later recognition they focus their inspection on particular views or perspectives; these same views are then preferentially utilized during recognition.

There is, however, an important limitation to these findings. It is that subjects in preferential viewing studies have not been free to select any viewing angle of a 3-D object by moving about it while at the same time being able to rotate it, behaviors typical of everyday interactions with objects. This paper describes a noninvasive, relatively inexpensive, and simple to use setup designed by the authors to record the visual inspection behaviors of individuals who are free to interact with the full view potential of a 3-D object, natural viewing conditions not found heretofore in studies of preferential inspection. We also present results of a demonstration experiment that verifies the potential of the system to study 3-D form perception.

The system, illustrated in Fig. 1, has two principal hardware components. One of these is a dynasight sensor (Origin Instrument Company dynasight TM sensor) which is used to track in time, and space the position of an observer’s head as he examines an object. The observer wears an eyeglass frame (without lenses) to which a small reflecting dot is attached at a point between the eyes. (If an observer wears eyeglasses, the frame rests on top of them.) The dot reflects the signal from the optical tracker directly back to it thereby measuring the position of the observer’s head relative to the center of the platform on which the stimulus object rests. The signal from the dynasight (x-, y-, and z-coordinates) is fed directly into a computer at a sampling rate of 10 measurements every second.

Fig. 1. The experimental setup showing the (1) participant, (2) turntable, (3) dynasight, (4) video camera, and (5) computer.

Objects are presented individually for viewing in the center of a round turntable mounted on top of a stable steel frame, as shown in Fig. 1. The turntable can be rotated 360° and beyond in either direction about the vertical axis. A gas-spring in the steel frame makes it possible to adjust the height of the turntable for each participant so that it is aligned with his or her elbow. This results in a standard and comfortable position of participants’ hands as they rotate the turntable while inspecting an object. After adjusting the height of the turntable for a participant, the height of the dynasight is also adjusted to the table. This is necessary to ensure that the participant’s head is located within the range of the tracker during exploration. As shown in Fig. 1, the dynasight and its fixture are located behind the turntable and in line with a participant’s face.

The second subsystem employed is a potentiometer (Bournes 3590S-2-502) which provides a continuous record of the location of the turntable relative to the observer as an object is inspected. The rotating shaft of the potentiometer is attached to the turntable and to a small electronic circuit. The computer controlling the setup reads head position values (through a serial interface) and table rotation values (through an Analog–digital converter), and attaches a time stamp to them resulting in five values per measurement. Data can then be analyzed in a world-centered frame (time and position values), or converted to an object-centered frame (by applying the table rotation to the position values). For the latter calculation, the position of the head position tracker relative to the center of the table must be measured and calibrated.

To examine participants’ hand and body movements as they examine stimulus objects, we have found it useful to make a video tape record of the entire experimental session. To this end, a video camera, positioned behind the dynasight and in front of the participant, is focused on the participant and the stimulus in such a way that the camera records the participant’s body and hand movements as he or she examines the stimuli.

The head-position data produced by the dynasight are combined with those generated by the potentiometer for a given trial to create a plot depicting the strategy used by an observer to inspect the stimulus object. The dynasight produces three orthogonal coordinates on a rectangular system. These coordinates are converted into a polar system, where the origin of the system coincides with the center of the turntable. Added to this set of data are those produced by the potentiometer during the same trial to create one inspection track around the object. By projecting these movements onto a virtual hemisphere around a stimulus, the resulting plot illustrates how a participant moved both the turntable and herself relative to an object as she inspected it. Examples of these plots are presented later in this paper.

This section describes a study which demonstrates the potential of our system for studying visual exploration of objects in 3-D space. The study examined the contribution of viewer- and stimulus-determined influences on the inspection strategies used by observers to learn abstract forms. Specifically, the participants’ task was to commit the appearance of six forms to memory, one form at a time, for later discrimination of each stimulus from a perturbed version of it.

Participants. 12 male and 12 female undergraduate students enrolled in the Industrial Design Engineering program at Delft University of Technology in Netherlands were paid a small amount for their participation.

Stimuli. The six pairs of abstract forms shown in Fig. 2 constitute the stimulus set. Each pair consisted of an original form sculpted by its artist and a structurally modified version of it made by removing some material from the edge of an exact replica of the original or by rounding one of its edges.