Ensemble coding of the average crowd speed using biological motion

Abstract

Perception of human crowds is integral to social understanding and interaction. Previous studies have shown that observers could accurately estimate the average facial expression, gender, family resemblance, joint attention, and heading direction of a crowd (for a review, see Whitney & Yamanashi Leib, 2018). The ability of extract summary statistics globally from a set of multiple features or objects has been term ensemble perception, a visual phenomenon which can occur rapidly and bypass the need to register local information of set members. The robustness and near automaticity of ensemble perception has sparked an ongoing debate as to how information of high-level social representations can be integrated. The parallel processing hypothesis proposes that multiple features are automatically integrated in parallel under distributed attention to form an ensemble percept (Treisman, 2006). The serial processing account argues that only a few set members are inspected in a serial manner within the capacity of visual memory and focused attention to compute summary statistics (Myczek & Simons, 2008). The current study examines ensemble perception of crowd speeds and addresses the parallel/serial processing debate using standard point-light walkers (Johansson, 1973). In the first experiment, it was found that observers could reliably and accurately estimate the average speed of a crowd. Ensemble processing of crowd speed could rely on local motion alone, however, the global percept of the human form enhanced all aspects of performance including precision, accuracy, and response time. The second experiment established that ensemble speed could be formed under rapid viewing duration, although speed estimation of slow crowds was less reliable and accurate than that of fast crowds. The efficiency of crowd speed perception was examined using an ideal observer analysis. This simulation revealed that only 2-3 walkers were integrated in rapid ensemble processing of slow crowds while 4-5 walkers were integrated in rapid ensemble processing of fast crowds. Together, these findings suggest that perception of average crowd speed may be a hybrid case of parallel and serial processing mechanisms and the degree to which observers can process high-level representations in parallel may depend on the spatiotemporal properties of actual stimulus speed. The study contributes to the current literature of ensemble processing of complex social characteristics, as well as provides the basic understanding in crowd speed perception. Such knowledge can be valuable to several real-world applications in public planning and monitoring of human crowds.