Skip to main content

Advertisement

Neural model of biological motion recognition based on shading cues

Article metrics

  • 319 Accesses

Point-light or stick-figure biological motion stimuli, due to the absence of depth cues, can induce bistable perception, where the walker is perceived as heading in two alternating directions [1, 2]. Psychophysical studies suggested an importance of depth cues for biological motion perception [3]. However, neural models of biological motion perception so far have focused on the processing of features that characterize the 2D structure and motion of the human body [4, 5]. We extend such models for the processing of shading cues in order to analyze the three-dimensional structure of walkers from monocular stimuli.

Model

As extension of a learning-based neural model [4], we add a 'shading pathway' that computes the internal contrast gradients that vary with the 3D view of the walker, even if the silhouette information remains identical (Figure 1A-C). The model exploits physiologically plausible operations. After suppression of strong external luminance gradients caused by the boundaries of the silhouette, internal luminance gradient features are extracted by a hierarchy of neural detectors. These gradient features, combined with the shape features extracted by the form pathway of the model in [4], are used as input for 'snapshot neurons', RBF units that detect 3D body shapes (Figure 1D). These model neurons are embedded within a two-dimensional recurrent neural field [6] that jointly represents the sequential temporal structure of the stimulus and the view of the walker.

Results

The neural field dynamics reproduces perceptual multi-stability and spontaneous perceptual switching between stimulus views, observed for silhouette stimuli in psychophysical experiments [1, 2]. It also reproduces the disambiguation by addition of shading information and a new perceptual illusion, which illustrates a lighting-from-above prior in the processing of biological motion stimuli.

Figure 1
figure1

A. Snapshot from a walker stimulus, rendered from a -45° side view. Vectors indicate internal luminance gradients, extracted by the internal gradient detectors of the model. B. Silhouette stimulus without shading cues is ambiguous and compatible with view angles ±45°. C. Snapshot and internal shading gradients for +45° side view. D. 'Shading pathway'. After suppression of strong boundary gradients, internal luminance gradients are extracted, using a hierarchy of neural detectors similar to a convolutional network. At the highest level is formed by 'snapshot neurons', RBF units that have been trained with keyframes from 3D walker movies, which are embedded in a dynamic neural field.

References

  1. 1.

    Vanrie J, Dekeyser M, Verfaillie K: Bistability and biasing effects in the perception of ambiguous point-light walkers. Perception. 2004, 33: 547-560.

  2. 2.

    Vangeneugden J, De Mazière P, Van Hulle M, Jaeggli T, Van Gool L, Vogels R: Distinct mechanisms for coding of visual actions in macaque temporal cortex. J Neurosci. 2011, 31 (2): 385-401.

  3. 3.

    Vanrie J, Verfaillie K: Perceiving depth in point-light actions. Perc Psychophys. 2006, 68 (4): 601-612.

  4. 4.

    Giese MA, Poggio T: Neural mechanisms for the recognition of biological movements and action. Nat Rev Neurosci. 2003, 4: 179-192.

  5. 5.

    Lange J, Lappe M: A model of biological motion perception from configural form cues. J Neurosci. 2006, 26: 2894-2906.

  6. 6.

    Amari S: Dynamics of pattern formation in lateral inhibition type neural fields. Biol Cyb. 1977, 27: 77-87.

Download references

Acknowledgements

Supported by EC FP7 ABC PITN-GA-011-290011, HBP FP7-604102, Koroibot FP7-611909, COGIMON H2020-644727, DFG GI 305/4-1, DFG GZ: KA 1258/15-1, and BMBF, FKZ: 01GQ1002A.

Author information

Correspondence to Leonid A Fedorov or Martin A Giese.

Rights and permissions

This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Keywords

  • Neural Model
  • Biological Motion
  • Neural Field
  • Motion Recognition
  • Gradient Feature