Low-cost Polling of Large Audiences

Electronic response systems known as “clickers” can enrich interactions in large audiences and have demonstrated educational benefits in well-resourced classrooms, but are cost-prohibitive in most environments. In an accompanying paper [1], we propose a new, low-cost technique that utilizes computer vision for real-time polling of large audiences. Our approach allows a presenter to ask a multiple-choice question. Audience members respond by holding up a qCard: a sheet of paper that contains a printed code, similar to a QR code, encoding their student IDs. Audience members indicate their answers (A, B, C or D) by holding the card in one of four orientations. Using a laptop and a digital camera, our software automatically recognizes and aggregates the audience’s responses and displays them to the presenter. In this supplementary note, we describe how to scale our system from classrooms of 25 students to large audiences of hundreds of people. At the 2012 ACM UIST conference, we conducted a poll with about 270 participants. Our system read 90% of responses, decoding them with 98% accuracy. INTRODUCTION Presenters facing large audiences often seek to engage and understand the personal views of their audience. A common requirement in many interactive and collaborative environments is the ability to poll students or audience members regarding their views or comprehension of a subject. One method of polling uses electronic response systems, in which networked devices called clickers are distributed to participants, allowing them to submit answers to multiple-choice questions (or occasionally, to submit richer data). Audience member responses are automatically aggregated and displayed to the presenter in real-time. In education, despite their benefits when paired with appropriate pedagogy, clickers remain out of reach for the vast majority of educational institutions due to their high cost. For example, one version of “clicker” called the i>clicker costs about $30 for each individual handset, plus $200 for a central receiver. For a university course with 100 students, even if a classroom already has a computer, an additional $3000 is needed to equip students with clickers, a cost often passed along to the students themselves. 1 In an accompanying paper [1], we describe an approach to audience polling that maintains the benefits of clickers while drastically reducing costs. Our system enables presenters to ask a multiplechoice question to their audience and receive their feedback without individual active components or a costly external receiver. Participants respond by holding up a qCard: a sheet of normal paper that has a printed code, similar to a QR code. The code * This is a self-contained supplement to our full paper, “Low-Cost Audience Polling Using Computer Vision”, which appears in UIST 2012. indicates the student’s ID, while the rotation of the card indicates the student’s answer. Using a computer vision algorithm and a camera with high enough resolution to capture the audience, the system aggregates the audience’s responses for immediate evaluation by the presenter. DEMO: POLLING LARGE AUDIENCES To demonstrate the capability of polling large audiences, we polled the entire audience at the opening session of UIST 2012. Of the 270 people who chose to participate in the poll, the system read 90% of responses, decoding them with 98% accuracy. During the break before the session, we passed out the cards on every chair. The front of each card had the code; the back of each card had a number associated with the card, and indications for which orientations corresponded to A, B, C, and D. During our talk, audience members were instructed on how to use the cards. They were told to hold the cards by pinching the top corners in the orientation corresponding to the intended response. Because of the density of people in the room, they were instructed to hold it at face height so as to be seen by the camera but not to obstruct the people behind. The audience was asked 7 questions in total including two ‘stock’ questions in which they were asked to hold up A, and then B, both for practice and to calibrate the system. Then, we asked the audience how many UIST conferences they had attended, three trivia questions, and finally a question regarding their preferred format for the next UIST. To capture the responses, one person stood on a 12-foot ladder with a Canon 550D DSLR camera. The reason for the ladder was to minimize obfuscation of cards by people or by other cards. In most university lecture halls, stadium seating would solve this problem and a ladder would not be necessary. For each question, five 18MP pictures were taken of the audience: one of each side of the room without zoom, followed by three zoomed-in pictures of the back left, middle, and right of the room. Each picture was processed at two different binarization thresholds because of variable lighting. Each picture was decoded independently and the results aggregated for display. If a card with a given identifier was decoded differently in two pictures, it was discarded from the analysis. RESULTS Overall, the demonstration was successful and the audience was very engaged with the presentation. The aggregated results of each poll are shown as an appendix to this document. Among other facts, our polls revealed that over half of the UIST audience was attending the conference for the first time, while 50 people Andrew Cross Microsoft Research India [email protected] Edward Cutrell Microsoft Research India [email protected] William Thies Microsoft Research India [email protected]