Limits...
Modelling human visual navigation using multi-view scene reconstruction.

Pickup LC, Fitzgibbon AW, Glennerster A - Biol Cybern (2013)

Bottom Line: Participants viewed a simple environment from one location, were transported (virtually) to another part of the scene and were asked to navigate back.We also measured error distributions when participants manipulated the location of a landmark to match the preceding interval, providing a direct test of the landmark-location stage of the navigation models.Models such as this, which start with scenes and end with a probabilistic prediction of behaviour, are likely to be increasingly useful for understanding 3D vision.

View Article: PubMed Central - PubMed

Affiliation: School of Psychology and Clinical Language Sciences, University of Reading, Reading, RG6 6AL, UK. l.c.pickup@reading.ac.uk

ABSTRACT
It is often assumed that humans generate a 3D reconstruction of the environment, either in egocentric or world-based coordinates, but the steps involved are unknown. Here, we propose two reconstruction-based models, evaluated using data from two tasks in immersive virtual reality. We model the observer's prediction of landmark location based on standard photogrammetric methods and then combine location predictions to compute likelihood maps of navigation behaviour. In one model, each scene point is treated independently in the reconstruction; in the other, the pertinent variable is the spatial relationship between pairs of points. Participants viewed a simple environment from one location, were transported (virtually) to another part of the scene and were asked to navigate back. Error distributions varied substantially with changes in scene layout; we compared these directly with the likelihood maps to quantify the success of the models. We also measured error distributions when participants manipulated the location of a landmark to match the preceding interval, providing a direct test of the landmark-location stage of the navigation models. Models such as this, which start with scenes and end with a probabilistic prediction of behaviour, are likely to be increasingly useful for understanding 3D vision.

Show MeSH

Related in: MedlinePlus

Likelihood maps for the four example conditions of Fig. 2, made using the “basic” map model of Sect. 4. Condition d, with particularly high angular uncertainty, is well captured by this model, but the more “radial” distributions of a and c are poorly explained
© Copyright Policy - OpenAccess
Related In: Results  -  Collection


getmorefigures.php?uid=PMC3755223&req=5

Fig4: Likelihood maps for the four example conditions of Fig. 2, made using the “basic” map model of Sect. 4. Condition d, with particularly high angular uncertainty, is well captured by this model, but the more “radial” distributions of a and c are poorly explained

Mentions: Figure 4 illustrates the generation of end-point likelihood maps using the data shown in Fig. 2. The model parameters were set to plausible values: 20 cameras spaced along a line 80 cm in length (i.e. matching the width of the viewing zone), and an “image noise” standard deviation of 0.05 given a focal length of 1m. 480 data points (10 from each of the 48 conditions described in Fig. 1) from a single participant were used to learn an optimal value for the parameter determining “decay rate” in the Bhattacharyya distance comparison. How well the model fitted these data is described later (Fig. 11), but for illustrative purposes, the fitted model is shown in Fig. 4 in the example conditions from Fig. 2. Our assumption in using the example points for illustration is that both these and the main set of 480 data points are sampled from the same underlying distribution. Note that the elongated “radial” distributions of end-points for cases (a) and (c) are not captured well by this model, though the more loosely clustered points in the other two cases are better explained. In Sect. 5, we will explore a modified version of the 3D model that is better able to account for this pattern of errors.Fig. 4


Modelling human visual navigation using multi-view scene reconstruction.

Pickup LC, Fitzgibbon AW, Glennerster A - Biol Cybern (2013)

Likelihood maps for the four example conditions of Fig. 2, made using the “basic” map model of Sect. 4. Condition d, with particularly high angular uncertainty, is well captured by this model, but the more “radial” distributions of a and c are poorly explained
© Copyright Policy - OpenAccess
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC3755223&req=5

Fig4: Likelihood maps for the four example conditions of Fig. 2, made using the “basic” map model of Sect. 4. Condition d, with particularly high angular uncertainty, is well captured by this model, but the more “radial” distributions of a and c are poorly explained
Mentions: Figure 4 illustrates the generation of end-point likelihood maps using the data shown in Fig. 2. The model parameters were set to plausible values: 20 cameras spaced along a line 80 cm in length (i.e. matching the width of the viewing zone), and an “image noise” standard deviation of 0.05 given a focal length of 1m. 480 data points (10 from each of the 48 conditions described in Fig. 1) from a single participant were used to learn an optimal value for the parameter determining “decay rate” in the Bhattacharyya distance comparison. How well the model fitted these data is described later (Fig. 11), but for illustrative purposes, the fitted model is shown in Fig. 4 in the example conditions from Fig. 2. Our assumption in using the example points for illustration is that both these and the main set of 480 data points are sampled from the same underlying distribution. Note that the elongated “radial” distributions of end-points for cases (a) and (c) are not captured well by this model, though the more loosely clustered points in the other two cases are better explained. In Sect. 5, we will explore a modified version of the 3D model that is better able to account for this pattern of errors.Fig. 4

Bottom Line: Participants viewed a simple environment from one location, were transported (virtually) to another part of the scene and were asked to navigate back.We also measured error distributions when participants manipulated the location of a landmark to match the preceding interval, providing a direct test of the landmark-location stage of the navigation models.Models such as this, which start with scenes and end with a probabilistic prediction of behaviour, are likely to be increasingly useful for understanding 3D vision.

View Article: PubMed Central - PubMed

Affiliation: School of Psychology and Clinical Language Sciences, University of Reading, Reading, RG6 6AL, UK. l.c.pickup@reading.ac.uk

ABSTRACT
It is often assumed that humans generate a 3D reconstruction of the environment, either in egocentric or world-based coordinates, but the steps involved are unknown. Here, we propose two reconstruction-based models, evaluated using data from two tasks in immersive virtual reality. We model the observer's prediction of landmark location based on standard photogrammetric methods and then combine location predictions to compute likelihood maps of navigation behaviour. In one model, each scene point is treated independently in the reconstruction; in the other, the pertinent variable is the spatial relationship between pairs of points. Participants viewed a simple environment from one location, were transported (virtually) to another part of the scene and were asked to navigate back. Error distributions varied substantially with changes in scene layout; we compared these directly with the likelihood maps to quantify the success of the models. We also measured error distributions when participants manipulated the location of a landmark to match the preceding interval, providing a direct test of the landmark-location stage of the navigation models. Models such as this, which start with scenes and end with a probabilistic prediction of behaviour, are likely to be increasingly useful for understanding 3D vision.

Show MeSH
Related in: MedlinePlus