Understanding intentions in animacy displays derived from human motion

摘要

As humans we live in a world where we are constantly interacting with those around us. To achieve this we must be able to successfully anticipate the intentions of others by correctly interpreting their movements. In studying how humans interpret intention from motion, we make use of simplified scenarios known as animacy displays where it has been shown that observers will attribute human-like qualities to the motion of geometric shapes (Heider and Simmel, 1944). This thesis advances the research into the attribution of social intentions by re-addressing the methods for the creation of animacy displays, leading to previously unexplored avenues of research. Where animacy displays are normally made via clever animations or mathematical algorithms, we introduce a method for creating these displays directly from video recordings of human motion, there by producing the first examples of animacy displays that are truly representative of human motion. Initially, explorative steps were taken to establish this technique as successful in creating displays that will be perceived as animate, using video recordings of simple and complex human interactions as a basis. Using a combination of tasks, including free response tasks and 10 point Likert scales, the use of this technique for stimulus production was validated. Furthermore, results showed that the viewpoint from which animacy displays are to be perceived from, comparing a side view and an overhead view, has effects on the ability to judge intentions in the displays, with a clear preference to the elevated viewpoint. Following this, the intentions of Chasing, Fighting, Flirting, Following, Guarding and Playing, thought to be generic to animacy displays, were used to create displays via this new method of stimulus production. Using a six Alternative Forced Choice (AFC) task it was shown that participants are successful at recognising these intentions, however, that the addition of ordinal depth cues, as well as cues to identity and boundaries, has little impact on increasing the ability to perceive intentions in animacy displays. Next, an experiment on the ability to judge intentions in animacy displays of brief durations was performed. Using the same 6 intentions as before, displays were created lasting 1, 5, and 10 seconds. Results of a 6 AFC task showed that observers are accurate at all durations, and furthermore, results indicate that participants are as accurate at recognising the intention in a display after 5 seconds, as after viewing longer durations of approximately 30 seconds. We then perform a comprehensive analysis of the animacy displays used, looking at the motion patterns and the kinematic properties such as speed, acceleration and distance of the agents. This analysis shows clear differences in the displays across viewpoints, and across intentions, that are indicative of the cues that participants may use to differentiate between intentions. We also perform a stepwise regression analysis to find the motion and positional predictors that best explain the variance in the behavioural data of previous experiments in this thesis. It is found that speed and acceleration cues are important for the classification of intentions in animacy displays. Finally, a study is presented that attempts to advance research into the perception of social intentions by people with Autistic Spectrum Disorders (ASDs), using video recordings of human motions and the resultant animacy displays. The intentions of Chasing, Fighting, Flirting, Following, Guarding and Playing, were again used in conjunction with a 6 AFC task. Comparing people with ASDs to an age-matched control population, results indicate that people with ASDs are poorer at judging intentions in animacy displays. In addition, results reveal an unknown deficit, not seen in the control population, in judging intentions from an elevated position in video displays. This work may be considered of interest to various groups of people with a wide range of research interests, including the perception and cognition of human motion, the attribution of social intent and “Theory of Mind”, and the surveillance of people via video techniques.