Abstract
Purpose: Human observers easily extract information (e.g., gender and identity) from point-light biological motion sequences. A common assumption is that point-light displays are highly impoverished relative to full-figured bio-motion. Our ability to easily perceive point-light sequences is then taken to suggest highly efficient use of available information. We used ideal observer analysis to test these assumptions by a) quantifying the relative information contained in full-figured and point-light biological walker displays and b) determining how efficiently human observers use the available information contained in each of these displays.
Methods: In two conditions, observers discriminated between a left and a right walking bio-motion sequence. In the FULL condition, bio-motion sequences were depicted by a dark silhouette of an actor walking on a treadmill. In the POINT condition, bio-motion sequences were depicted by a point-light walker stimulus that was generated by tracking 13 body coordinates from the FULL walker sequences. Gaussian spatiotemporal pixel noise was added to the walker and discrimination thresholds were obtained by varying the walker contrast. FULL and POINT thresholds were measured in separate blocks, with the order randomized across subjects. Ideal observer thresholds were measured for the same two conditions.
Results & Conclusions: Surprisingly, ideal observer performance was indistinguishable in the FULL and POINT conditions, indicating that each carried the same amount of discriminative information. However, human performance was up to 7 times worse in the POINT than the FULL condition, indicating that human observers used information far less efficiently in the point-light displays. Efficiency (ideal/human threshold) ranged between 0.4–0.6% in the POINT condition and 2.2–3.0% in the FULL condition. We are currently using response classification analysis to investigate this relatively inefficient use of information in point-light displays.