Unfortunately, I missed the first talk of this session.
3D Tracking via Body Reflections
- Want ability to track people’s movement without requiring them to carry a device. Approach is to broadcast radio signals and measure reflections
- Can we see through walls with wireless signals?
- By shining wireless signals on walls, can one see people moving behind wall?
- Video demo shows 3d tracking working in a variety of situations. Pretty impressive.
- Centimeter-granularity tracking when used in non-obstructed situations
- Applications: gaming, gesture control, first responders, etc.
Approach & how it works
- There is a minor reflection between person behind mall. Need to measure it. Distance = reflection time * speed of signal (light).
- Option 1: Transmit pulse and listen for echo. But, need mult-ghz samplers, which are impractical.
- Use frequency modulated carrier wave. Transmit signal where signal linearly increases with time. Received signal will be a shifted version of transmitted signal. Turns out transmitted and received signals are also shifted with regards to frequency. reflect time = <>frequency/slope of signal.
- Measuring change in frequency is easy, can be done using a ‘mixer.’ Apparently, this is cheap.
- transmitted signal/received signal -> mixer -> fft -> shifted frequency
- One problem: multi path. One transmitted signal will have many reflections, not just the person you want to track
- Can deal with static objects because they don’t move. Subtract consecutive measurements
- Even removing reflections from static objects can result in multiple signals fora single person. This is because person’s reflection off of static objects will also change as person moves. Can deal with this by only selecting signals coming from shortest distance
- Needs two antennas [missed the reason for this]
Implementation & evaluation
- Comparison to ceiling mounted IR transmitters that can sub-locate very accurately
- One experiment shows whether WITrack can measure person’s orientation. Median error is11 degrees, which is really good.
- One application: Fall detection. E.g., did an elderly person fall and someone be called to aid them?
- Graphs of elevation vs. time for fall detection show clear differences for walking, ersus sitting on a chair, versus lying on a ground (falling). Accuracy of fall detection is 97%
- 3D tracking also allows tracking breathing activity
Epsilon: A Visible Light-Based Positioning System
- Lots of need for precise tracking, for example, finding books on bookshelves, etc. Wi-fi based tracking has accuracy of only a few meters
- LEDs are becoming more common place. They have 4 microsecond on/off time.
- Lots of other people use LEDs for visible light communication. Want to also use it for positioning
Constraints and implementation
- When incidence angle of light against objects increases, error increases
- LED highs must have certain specifications (support up to 100kHz frequency and have lower bound greater than 200Hz)
- Light samples on commodity phones can only sample light within a limited range
- Prototype modified off-the-shelf LED. Added a control board implementing BFSK beaconing.
- Phone had an extra light sensor attached via the audio jack
- 90% of errors in positioning are less than 1 meter. Achieves very close results to best wi-fi method
- All of this depends on an infrastructure that doesn’t exist. Are we going to all need smart light bulbs? Answer: [I didn’t quite get this]
- What are privacy implications? Can you compare privacy risks to this to other systems? Answer: There’s a dimension other than accuracy. This can’t see through walls
- Would using a higher frequency help? Answer: No help in accuracy
- If you have a lot of LEDs instead of just one, can location accuracy improve? Answer: Yes, it will improve, but don’t know how much
- How does orientation of phone affect numbers? Answer: With multiple light source, it won’t matter. If only one, will really affect accuracy.