Motion-Captured Locomotion

>>Download database here<<

  • All stimuli in this study were locomotory actions, i.e. displacement of a human body, but whereby the translational component is removed, thus resembling an actor as if locomoting on a treadmill. This was done in order to more accurately pinpoint the neural code subserving walking direction (walking left- or rightward either in a forward or backward fashion), for which actual physical displacements of the body are erroneous in determining the neural code.
  • Monkey subjects were instructed to discriminate either forward from backward actions (i.e. forward-backward task) or different facing directions (e.g. facing to the left or to the right; i.e. view task). We generated numerous versions of each action by varying the starting position. Besides these actions locomoting at a normal pace of 4.2 km/h, which the monkeys subjects were extensively trained on, we also created similar actions, but at different speeds (2.5, 6, 8, 10 and 12 km/h: ‘02-prefixes’). These actions at different speeds were created in order to check how general the learned discriminations were. If the monkey subjects rather learned a general abstract concept of ‘walking forward to the right’ etc., than we would expect quite good discrimination performances for actions of the same category (e.g. ‘forward left-2-right’) even though the speed is not equal.
  • A same logic applied when assessing actor-invariance and tests for the generalization of stimulus versions. We also generated actions containing just parts (upper or lower body) of the original full body configuration. This was done in order to examine which part of the body configuration contributed to each task separately. Finally, we also monitored discrimination performances when gradually reducing information present in the stimulus, i.e. applying spatio-temporal scrambling. The ‘percentages’ refer to the level of noise added to the walker. Every 5th frame, we reshuffled the scrambling pattern ane.
  • Purely at the temporal level, thus without tampering in the spatial domain, we developed a few new stimulus types. In one version we removed frames at certain intervals and replaced them by blank frames. This gives the impression of a stroboscopic experience (‘08-prefix’). In a similar manner, instead of replacing frames by blank frames, we froze the frame prior for a number of frames, before jumping to the next frame in the original action (’07-prefix’). The study was concluded by controlling for opponent motion. We generated a number of variant locomotions containing no opponent motion (‘09-prefix’).

Vangeneugden J, Vancleef K, Jaeggli T, Van Gool L, Vogels R (2010). Discrimination of locomotion direction in impoverished displays of walkers by macaque monkeys. Journal of Vision, 10:22.1-22.19.

  • This study was set up to investigate how temporal cortical neurons encode actions differing in direction, both forward versus backward as different facing directions. Computationally, coding between forward and backward versions differs with respect to which aspects of the actions are relevant compared to coding for different facing directions. We therefore motion-captured an actor walking at a normal pace on a treadmill, from multiple camera angles: in the lateral (0° and 180°), frontal (90° and 270°) and oblique planes (45°, 135°, 225° and 315°). All possible combinations amount to 16 actions in total. Also, for each action, we extracted a number of static frames (i.e. snapshots) representative of that action (‘static_poses’-postfix). Finally, we also generated upper-body and lower-body only action variants (‘UppBody’ - and ‘LowBody’ - postfixes).

Vangeneugden J, De Mazière P, Van Hulle M, Jaeggli T, Van Gool L, Vogels R (2010). Distinct mechanisms for coding of visual actions in macaque temporal cortex. The Journal of Neuroscience (in press).

  • All the stimuli are available as *.avi such that playback is not limited to certain devices.