Cell phones can monitor the goings-on at parties and shoot video highlights using software developed by researchers at Duke University.
VUPoints collaboration software organizes cell phones carried by party goers into affinity groups, determines when something interesting is happening and tells the phones with the best vantage points to capture the moments with their cameras.
The software uses sensory input from the phones to figure out which of them are close to each other and therefore are probably carried by people who are interacting, such as a group sitting at the same dinner table at a banquet. Researchers presented their work at the recent Sigcomm conference in Barcelona.
The researchers had four people with Nokia N95 phones taped near their shirt pockets interact with each other to stage an event. The phones had built-in video cameras and accelerometers. The event was recorded in its entirety by a separate phone, and a person unfamiliar with the experiment viewed that recording and identified highlights. Those highlights were compared with the ones determined and recorded automatically by VUPoints.
For one particular five-minute test event, the human viewer identified a minute and a half of highlights. VUPoints identified two and a half minutes, which included all of the events the human viewer highlighted plus another minute of false positives.
Over four experiments, VUPoints recorded 80% of the interesting moments as identified by a person who watched the whole event. The average false positive rate was 33%.
VUPoints relies on a server that is fed data from the phones and analyzes it. Similar lighting and sound activity indicate phones that should be in the same social group. Also, the server can trigger one phone to emit a ringtone above audible frequencies and other phones respond with how well they heard it. Those that hear it at about the same level are grouped together.
Alternatively, phones can all submit the ambient sound they hear at a given moment, and by comparing the samples, the server can figure out which phones seem to be clustered. The server can then assign phones to socio-acoustic groups. The server could determine events of interest by noting that several cameras are looking at the same object. When it determines similarity among images from group phones, it triggers video recording by the group.
