3. CONTEXTUALIZACIÓN DE LA EXPERIENCIA SISTEMATIZADA
3.2. Trabajo Social en la Universidad del Valle
To actually define interest, the actions, posture, and head pose of the students from Chapter5 are all considered in the definition that follows. Images where a subject is present and not occluded are labelled as valid (1) – only these valid images are considered and images that classified as invalid (0) are automatically excluded. Let the label ‘Interest’ indicate that a
student is interested or engaged, and ‘¬Interest’ indicate that the student is not interested or disengaged. The assumptions listed below are based on the literature surveyed in Chapter 3 as well the experience developed during data labelling and lecturing in general. Limitations are discussed where appropriate. The assumptions that follow form a sieve – the rule either causes a student to be marked (¬)Interested or it has no effect. If a student passes through all the rules without being explicitly marked one way or another, then the student is considered to be engaged.
There are seven actions identified in the database: writing, cellphone use, laptop use, talk- ing, raised hand, yawning, and head-on-desk. These are defined in Section5.3.5and illustrated in Figure5.9.
The following rules about actions are considered:
Writing⇒ Interest Cellphone⇒ ¬Interest Talking⇒ ¬Interest Raised Hand⇒ Interest Yawning⇒ ¬Interest Head on Desk⇒ ¬Interest
The fundamental assumption is that a student that is writing is actually taking notes and not just drawing or playing ‘Dots and Boxes’ with a friend. A student with a raised hand is either asking a question or responding to one and therefore engaging with the lecturer. On the other hand, a student that is on a cellphone or that has put their head on the desk is considered to have disengaged and lost interest. Yawning is considered to be a symptom of fatigue that ultimately leads to disinterest. A student talking to a friend may be discussing work, asking a question, or talking about an upcoming party. WITS is unable to tell the difference between these cases, so the assumption is that they are just chatting and therefore disengaging. It is unclear whether laptop use implies interest or not. In many real-life cases the students are taking notes, and in many others the students are playing games or browsing the internet. As only one subject in the dataset used a laptop, for the purposes of this research the debate can be avoided and laptop use is ignored completely.
There are five labelled upper body postures that correspond to the images shown in Figure 5.11. Specifically, these postures correspond to leaning left, right, backwards, forwards or sitting upright.
The following rules about posture are considered: Left⇒ ¬Interest Right⇒ ¬Interest
A student that is leaning far to either side is always disengaged. This behaviour co-occurs with talking to friends, sleeping and cellphone use. Those leaning only moderately to the side still exhibit a closed body posture that is identified in literature to occur with combative (resistance and aggression) or fugitive (withdrawal, defensive, boredom) stances. In both cases, this implies disinterest or disengagement on the part of the student. Those students with an
open posture are usually sitting upright, leaning forwards or backwards. Whether this indicates interest or not, largely depends on a student’s head pose and actions. Therefore, there are no rules that specifically deal with these poses, but if the student is not excluded in the steps that follow, they will still be marked as interested.
For example, in Figure 5.11b on page 110, the student is leaning back while watching the lecturer with an open body position and a frontal head pose – this student is evidently concentrating on the class. Figure5.12c, on the other hand, shows a student leaning back so that he can see the cellphone under the desk – in this case the student is obviously disengaged from the class and trying to hide it. In the latter case, the student will pass through the rules relating to body posture, but will be excluded by either the cellphone use or because is head pose focuses his attention under the desk.
The head pose labels are illustrated in Figure 5.12 and mostly corresponds to a student’s focus of attention – unfortunately the resolution of the images does not allow for eye tracking. The horizontal labels correspond to far left or right, moderately left or right, or forwards. Ver- tical labels correspond to looking below the desk, on the desk, forwards, or upwards. These labels lead to the following rules involving head pose:
Far Left⇒ ¬Interest Far Right⇒ ¬Interest Up⇒ ¬Interest Below Desk⇒ ¬Interest
Students looking to the far left, far right, or up at the roof are not focusing on the lecturer or their notes. They are either looking at a friend, or looking around the lecture venue. Sometimes it is the case that they are focusing on another student that is asking a question. Detecting this would require a global consideration of the students in the video and should be the subject of future work. Those that are looking below the desk are almost always distracted by something that they are trying to hide from the lecturer – usually a cellphone as illustrated in Figure5.12c. Altogether these rules can be considered in a manner similar to the cascade classifier. Firstly, the Writing and Raised Hand labels are checked – if either is true, the student is marked interested. The remaining rules are all used to reject students and mark them as¬Interested. If a student successfully passes through all ‘weak rules’ without being rejected then the student is labelled as Interested.