Contextual Inquiry-Group:The A-Team
From CS160: User Interface Design Sp12
Project completed by the The A-Team.
- Shu-Chen Chen - Interviewed Professor X, wrote Competitive Analysis.
- Benjamin Coleman - Contacted and Interviewed GSI Z, wrote Analysis of Tasks, and contributed to Problem Solution and Overview, Task Analysis Questions, and User Interface Design.
- Chenkai Gao - Interviewed Professor X, contributed to Task Analysis Questions, Problem Solution and Overview, and User Interface Design.
- Jingwei Qi - Contacted and Interviewed Professor Y, contributed to User Interface Design.
- Raphael Townshend - Contacted and interviewed Professor X, wrote Target Users and Contextual Inquiry. Contributed to Problem Solution and Overview, and Task Analysis Questions.
New Project idea: We are switching our idea to Contact Detection and Management System.
Why: We were considering doing this idea to begin with, but ended up selecting the building use counter due to the need to select one idea. However, upon further consideration this idea will be more challenging, more fun, more useful, and it is relatively easier to access the user group
Who and Why
Computer Science instructors need to meet a lot of students and interact with many different people. Sometimes they can't remember critical details about the people they are interacting with. Thus, a contact management kinect application would be particularly well suited for them. We decided to focus our inquiry around office hour times, where instructors are very likely to interact with many students. However, we also considered alternatives during our interviews.
Instructors can generally be categorized as professors or graduate students.
Computer Science professor: PhD level education and a significant amount of teaching experience.
GSI: Bachelors or higher level education, with a least a basic amount of teaching experience.
Professor X -- Experienced Computer Science professor currently teaching upper division courses. Professor X meets a lot of people (undergraduates, graduates, professors, industry professionals). He is open to new technology, but wants an unobtrusive way to be assisted in remembering students' and GSIs' names. Does not like technology that would make social interactions awkward, unless said technology becomes generally accepted.
Professor Y -- Experienced Computer Science professor currently teaching upper division courses. Very open to new technologies and methodologies but very busy (research, meetings...) and does not have much time to commit to contact management.
GSI Z -- Third year PhD student currently TAing for upper division courses. Experienced in teaching and willing to put extra effort in to manage contacts. Unsatisfied with current options for managing contacts in the context of office hours. Less busy than professors but also have many different commitments to deal with.
Problem Solution and Overview
The problem: Instructors may not always remember the names and other critical details about people they interact with. For example, they often have the need to recall which students they interact with on a regular basis during office hours, or what research a colleague they meet at a conference is working on. This problem comes from a disconnect between human memory capabilities and the expectations others have in terms of being remembered.
The solution: Addressing the specific problem of office hours, we can use the Kinect to detect and identify people as they enter the room (using facial recognition and skeletal tracking for positioning), then publicly display their name along with their face and other details in a list of people present. Details of particular use to a professor may include the student's interests, the student's previous questions, and the pronunciation of the student's name (use the kinect's microphones to capture the student saying his own name). This data would be publicly shown in the office in order to create an implicit understanding and convention between all users that said data is available, thus avoiding any potential awkwardness.
Note on Format of Interviews
An actual in-depth contextual interview was quite difficult to obtain due to the nature of office hours, and thus a balance was sought by performing both a conventional interview, and in-context observations and short in-context interviews.
Conventional 'sit-down' interview with a walkthrough of concrete tasks performed during the professor's day. The master-apprentice model was used during the interview. Probing was performed on topics of interest mentioned by the professor during the interview, though the initial talking points were based off a set of questions seeking to challenge our assumptions. Interview was 30 minutes long and performed in Professor X's office.
Interview Summary and Individual Insights
- Meetings occur
- In office about 4 days in a week
- A day or two in silicon valley or in some coffee shop (less often nowadays because it takes too much time to go back to office)
- Terrible with names
- Information needed when at meeting
- Why he knows that person
- Information needed when meeting a student in OH
- How many times he met this student before
- The length of their previous conversation
- What they talked about from previous conversation (particularly useful when he meets grad students, they need to talk about the progress of their projects)
- Potential uses of the system
- Show the name
- How to pronounce the name
- How well does he need to know this person (including how many times do they meet before)
- The way which the system presents the information
- The system must show information only to the professor
- Hide itself from the students or the person the professor is talking to
- This is important because it is not friendly when the person who you are talking to find out that you don't know his name
- Additions to the system
- Link and integrate with Facebook and other social media can let the professor know what student is interested in
- Can probably remind him the previous conversations between professor and student
- Uses in a conference
- Might want to know information other than name and work place concerning another attendee, such as what research he is doing, and whether he has a talk coming up
Observations of professor during office hours followed by relevant questions concerning what was just observed.
Interview Summary and Individual Insights
- Let's students work on problems themselves to deal with high throughput of students
- Uses this system to try and reach many students
- Spends time evaluating solutions with each group of students
- Can be problematic to keep track of who still needs their questions answered
- A way to optimize office hours would be nice
- Professor can't remember students name very well, even if one student just told him his name several minutes ago.
- Sometimes the professor mismatches students and their names.
- Even though the professor remember the student's name, he has trouble pronouncing it.
- Both general answering and one-on-one answering happened.
- Use of kinect would be awkward when professor does general answering if the Kinect is not mounted on him.
- The professor doesn't have ways to remember names other than using his memory
- Too busy to be able to keep track of names
- Would like to remember which students have asked him which questions
- Lots of potential information available online that could be of use
Observations of graduate student instructor during office hours. Notes were taken during the GSIs interactions with students. Observation was followed by short contextual inquiry at the conclusion of the office hours.
Interview Summary and Individual Insights
- If the office hours are in a room like 611/651 soda, the students may be facing different directions.
- Problematic for facial recognition
- More 1-on-1 interaction than general question answering.
- Interactions tailored to specific student questions
- No general Q&A
- Different at a midterm review session
- Need to keep track of unanswered questions in certain settings
- Difficult to remember specifics of each student
- Remembering names is an issue
- The professor gets a pdf with names and mugshots that the gsi can request
- If the gsi wants to put names to faces he must spend time memorizing that pdf
- Students expect to be remembered
- Thought that our system would be socially awkward
- Types of questions asked
- Midterm review questions
- Homework questions
Insights on Common Tasks and Themes
All instructors have issues remembering names, pronouncing names, and remembering details of individuals they have previously met in office hours. They do not want to publicly reveal that they do not remember much about the person they are engaging. However there is also a feeling that the expectations concerning their own memories are unreasonable. An interpretation to draw from this is that current cultural norms impose an excessively demanding requirement on 'celebrity' type figures (i.e. people who meet many others regularly) to recall others, and that some sort of societal change is necessary to lessen this burden. Thus, a kinect application that would help change this 'social contract' would be very useful. One way to do so would be to make the obtention of information concerning the person you are engaging convenient, rapid, and most importantly public. The public aspect would create an implicit understanding between celebrity and 'paparazzi' that said method of obtaining information is socially acceptable.
Specifically in terms of office hours, they nearly always occur within a professor's office or another small room frequently used by the instructor. The ability to go over concepts in need of review and hold basic conversation with students is also important. Tracking who still has unanswered questions also would be an appreciated feature. The specifics of what an instructor wishes to remembering in terms of details of a student vary from instructor to instructor (how many times the student has been met before, what kind of questions they have previously asked...). All instructors attempt to attend to their students but the methods for doing so vary and are never fully effective. One-on-one interactions would be most desired but Q&A often occurs in a group format.
If the information is not make fully public, then the need is really to keep the information feedback private, unobtrusive, and discrete as it would otherwise render social interactions during office hours quite difficult and awkward. Students would be unsure what was going on, and that would generate uncomfortableness on both the professor's and student's parts. The potential awkwardness of using a kinect application for contact management was especially emphasized during interviews, and thus social conventions will have to play a big part in any interface we design.
Finally, there was a general desire to integrate online information (Professor X specifically mentioned Facebook) with a user's displayed information, though this may be a little complicated if the display is public. Perhaps a display of a student's pending questions would be a good idea.
Task Analysis Questions
1. Who is going to use system?
The users are professors and GSIs who wish to have easier ways to remember the names and other details.
2. What tasks do they now perform?
Currently professors must rely on their memories for both student names and past difficulties the students have had. They then use this to help the students during office hours.
3. What tasks are desired?
Using the Kinect to identify a student, and provide students personalized assistance. Professor X would like to know how many times he has interacted with the student beforehand.
4. How are the tasks learned?
They use official photos and spending hours study them in order to remember them better. GSI Z uses a pdf of student faces to attempt to remember names.
5. Where are the tasks performed?
The office hour and meetings usually happened indoors in a classroom, conference room, or office.
6. What’s the relationship between user & data?
The data is currently stored in the user's brain, or in some written form. We want to store it in a computer memory to speed up recognition and increase accuracy.
7. What other tools does the user have?
The user have smart phone, computer with web camera, papers with student name and photo on them which can also help them recognizing the names.
8. How do users communicate with each other?
Professor and GSIs communicate by meeting or emails. Office hours are publicly posted.
9. How often are the tasks performed?
The task are performed multiple times in a day during weekday. Recalling a person's name is especially done during office hours, conferences, and meetings.
10. What are the time constraints on the tasks?
The instructor may have few seconds to recognize the name, and office hours are of limited length. Professor Y attempts to solve this by splitting students into work groups.
11. What happens when things go wrong?
The instructor can get embarrassed, and the students can get confused. Professor X emphasized how awkward the situation becomes.
Analysis of Tasks
Recognizing faces and putting names to faces
During a professor or GSI's office hours, he may interact with anywhere from a handful of students to tens of students. These students may come out of a class of several hundred and may show up only once or may show up on multiple occasions. Although remembering names can be done, many professors of large lectures state that they will not remember names simply because there are too many people to keep track of.
Remembering background information
Some students may have issues with some concepts repeatedly. Knowledge of these is useful when providing one-on-one assistance.
Students will ask the GSI/Professor for help with specific issues during office hours. The professor/GSI then aids the student with the problem.
Some people have really hard to pronounce names and some professors/GSIs may be professional name butchers.
If it's not busy, some people may stop by and chat.
Reviewing general concepts
Sometimes people need help with these.
The basic interface we’re trying to make is a controlless passive interface. It will use the depth data to detect person when someone enters a room then grabs a few video frames to run facial recognition on.
- Scenario 1: Recognizing faces and putting names to faces.
After recognizing the people, it displays the information (name, picture, etc.) of them in a big screen or on a projector. It will rescale pictures, names and other info depending on the number of people present so as to make the best use of the screen real estate, but after a certain point it should stop shrinking information sizes and instead have scrolling function.
- Scenario 2: Students unknown to the system, store information.
When the system doesn't recognize the student's face, it'll display the picture and show "unknown" at the name section.
- Scenario 3: Student asking questions
When students have questions, they can perform some gesture before the kinect, the background color of that student will change to indicate they have questions so that the professor know which students have questions and the priority sequence (FCFS).
List of Competitors
· Kinect Facial Recognition and Transfer (http://www.kinecthacks.com/kinect-facial-recognition-and-transfer/)
· 3D FastPass (http://www.tis-uae.com/doc/L-1_3DFastPass_datasheet.pdf)
· 3D Facial Performance Capture using Kinect (http://www.youtube.com/watch?v=nYsqNnDA1l4&feature=player_embedded)
· SmartGate (http://en.wikipedia.org/wiki/SmartGate)
· Face.com (http://face.com/)
Since Kinect device hasn't been in the market for long so we can't get exactly the same usage applications as ours. However we tried our best and found the 5 applications which are as similar as our application.
Kinect Facial Recognition and Transfer
· Target User Group:
The target users of this software are 3D animation movie makers and game makers who would like to create characters with accurate facial expressions. However, our target users are those who would like to recognize who the people are in front of the camera.
The application dynamically captures the user’s facial expression using Kinect’s depth camera, construct a 3D facial model and then transfer the model to a model of the target character, while our idea is to identify the user using their facial features.
The user of this application can just sit in front of the Kinect and his facial expression is dynamically captured and transformed into the target character.
· Target User Group:
This application is an access control system. Its target users are companies or organizations who would like to control the access of their facilities and equipments. These are different from our initial target users, who are individuals who would like to identify people they contact.
The user needs to stand in front of the device and position them inside the video screen. The system then captures the user’s features and grants the user the access to the facility if he is recognized. This is similar to our application.
The user of this application just needs to stand in front of the camera. No typing or button pressing is needed. This matches our user scenario.
3D Facial Performance Capture using Kinect
· Target User Group:
This is just a research project, so no target users are mentioned in the website, but I think its target users would also be 3D animation movie makers and game makers like the first one.
The same as “Kinect Facial Recognition and Transfer”, the application dynamically captures the user’s facial expression using Kinect’s depth camera and construct a 3D facial model.
· Usability: The user of this application can just sit in front of the Kinect and his facial expression is dynamically captured and the 3D model is updated.
· Target User Group:
SmartGate, which was introduced by the Australian Customs and Border Protection Service and New Zealand Customs Service, is a border control system. Its target users are customs at the border of a country.
The application captures a live image of a person in the camera and matches it with the person’s passport picture. Therefore, the system actually does not try to identify who the person is. This is a bit different from our application. Besides, only 2D picture data is used in this system, while our system will use 3D facial data for recognition.
Since this system just report the degree of matches between the live image and the passport photo, no training data in the database is needed.
· Target User Group: face.com is a technology company operating the face recognition platform They focus on enabling social experience for people in images, and their platform is built to support web and mobile services, directly off the cloud. This app is for everybody who would like to improve his/her social networking experience.
The application offers a platform for developers and publishers to automatically detect and recognize faces in photos. It also has photo tagger and celebrity scan functions.
This web application allows third-party developers offering instant integration to social networks; photo sharing sites, search engines, and more.
Based on the information of all the reviewed competitors’ applications, Kinect’s depth data is either used to track a person’s facial expressions, or used to identify who the person is. However, for those applications that recognize people based on their facial features, their target users are usually those who do not need to social with the people being recognized. This is the main difference between those applications and ours and is probably worthwhile for us to rethink who our target users should be, since the recognition process and result could embarrass the user if he needs to have social relationship with the one being recognized. Anyway, we can clearly see the potential of facial recognition technology using Kinect since its 3D data improves the recognition accuracy.