In recent years there has been an increasing interest in several fields of research for analyzing human behaviour. Typically these works focus on macro-level recognition such as activities or explicit human-human interactions.
Systems built to address these tasks assume that what is recognized corresponds to the true intentions of the observed person. This may not be always the case, since macro-level behaviors can be easily counterfeited.
For this reason, the interest in understanding micro-level behaviors has emerged, since these are often connected to involuntary reactions which are less likely to be faked or concealed.
Understanding strong, unbiased behaviors is also a key aspect in several fields. For example, more accurate interest profiles could be built from facial micro-expressions, leading to better product or content suggestions.
In security critical applications, observing posture, gait and small body movements could reveal malicious intentions.
Moreover, body movements and behaviors could also be used to profile or re-identificate subjects without harming their privacy.
In general, low-level expressions and behaviors are more difficult to recognise than high-level ones, due to the fine-grained nature of the problem. It is also a fact that with the recent technological advancement, the means of data acquisition and processing have also dramatically improved, enabling new applications and analyses.
To reliably understand facial micro-expressions, for example, there are both spatial and temporal issues to take into account. On the one hand it is necessary to either acquire and process high-resolution images or rely on different kinds of data such as high-quality depth maps.
On the other hand, micro-expressions occur over an extremely short timespan (<500ms) and might not even be detectable with conventional low-framerate cameras. If some years ago it was not possible to acquire these kind of data and process them, now it certainly is.