Flashcards
L3: Sensors
Cards in this deck (57)
- 1What are the two fundamentally different approaches for ubiquitous interaction?One is focus on the environment, one on the user (what the user carries with them)
For example: Kinect - tracks proximity and therefore movements of the user
PS Vita - a user carries and plays with
ex2:
on user: a microphone that listens to the sound your bones make as your move your wrist/fingures and classify certain gestures
environment: kinect can recognize gestures using infrared - 2What is the ultimate goal of ubitiqous comp?Systems should be aware of all intent, desire, actions and activities of users
- 3What has driven the Hardware renaissance?▪ Rapid Prototyping - Arduino, rasper (microcontrollers) has made it possible for normal people to pick it up and create things, play with physical sensors and prototype their own ubiquitous computing systems.
▪ Mass Customisation - A lot of people have certain needs and are driven by a need to customise their own hardware
▪ 3D printing:
Rapid prototyping
Mass customisation
Cheap Deployment - 4What is Moores Law?Number of transistors per square inch on integrated circuits had doubled every year since their invention
- 5What is a sensor?A sensor is a device, module,
or subsystem which detects events or changes in its environment by translating a physical magnitude into electrical
properties. Helps us makes sense of the context/world that our applications live in.
You have some physical phenomena that has certain magnitude, attributes (pitch, frequency), scale, pitch, frequency which can be picked up by a sensor and translated into a signal. - 6Give examples of sensors on a Nokia phoneMicrophone, buttons, antenna, battery level
- 7What is a biometric sensor?A biometric sensor is a transducer that changes a biometric treat of a person into an electrical signal. Biometric treats mainly include biometric fingerprint reader, iris, face, voice, etc.
- 8What sensors are being used when you rotate your screen?Accelerometer
Gyroscope - 9What sensors are being used when you avoid accidental touchesproximity
- 10What sensors are being used when you play pokemon go?GPS, camera, touch (for user input)
- 11How can you measure which floor you are on in a building?Using barometer to sense air presure so you can estimate the height
Often used to refine navigation - 12What sensors are being used when you are driving through a tunnelDead reckoning: Accelerometer using to predict where you will be at a given time
- 13what is the IMU?Inertial measurement unit - an electronic device that measures and reports a body's specific force, angular rate, and sometimes the orientation of the body, using a combination of accelerometers, gyroscopes, and sometimes magnetometers.
- 14How does listening do sensors fit into the lifecycle of your app?subscribe to sensors on on Start(), start listening on sensors onResume()
unsubscribe listening on sensors
on onPause() and onStop() - 15How do we sample sensor datawe need to take discrete samples of each state, which is the frequency which we measure the state. The higher frequency measures - the higher sample - the more accurately we can reconstruct the original signal
- 16What kind of problem are you dealing with when discrete sampling?You loose data on peek.
You have a lot of noise - 17What is SNR?Signal to noise ratio.
The more noise you have, the lower signal to noise ratio
You want high SNR - 18What is the difference between linear regression/machine learning and signal sampling?In machine learning/linear regression you look at usually one example at the time.
But with sensor data is often a sequence, and each measurement is dependent on what happened before (for example when running). - 19Why is it important to filter signals?It we make decision on signals that are faulty - bad things can happen! We need to make sure that we can filter our noise and make good predictions.
- 20What are the five methods of sensor data processing?1.Thresholds
2.Mean Filters
3.Median Filters
4.Kalman Filters
5.Particle Filters - 21How do you measure signals with Treshold?As soon as the signal exceeds a certain threshold - you consider the signal to be present in your system. Could be just 0’s and 1’s. You can have multiple thresholds and you only consider the signal in between two thresholds.
- 22What kind of treshold filter can you use if you want your signal to be turned on at one treshhold but turned off at another?
- 23When are treshold used?automated volume regulation in earphones
- 24How is the sensor noise distributed?The offset vi is normally distributed
- 25Why is the measurement from the sensor rarely equal to the actual value / ground truth?Sensor noise
- 26How does the mean filter work?We take a window
- 27Pros and cons with mean filterPositive: easy to calculate
Negative: susceptible to outliers and you need to wait for the window to fill which means you get a lag. In addition, makes up data as it calculates the mean, don't use the actual measurements - 28What happen when you increase window size in mean filter?The outliers get less bad (but they are still there)
The lag becomes worse
The sharp edges also get smoothened out but it works well for straight lines. You miss you miss details such as the zig-zag patterns. - 29One problem with mean is that each measurements has the same “value”. How can you improve it?Add Weighted mean. We assign higher weights to the ones that are more recent (for example)
- 30Why does the median filter deal better with outliers than the mean one?It only picks the middle value, which mean outliers are usually not counted
- 31Pros and cons with median filterPro: Less susceptible to outliers.
Doesn't ‘make up’ data. Each point in my filtered data has been measured.
Deals better with zig-zag patterns and edges than the mean
Con: has to wait for the window to fill - aka lag - 32Pro/cons with median & mean filterMean/Median Filter summary:
+ easy to implement
+ efficient (the window size is all your have to keep in memory)
+ great cost-benefit (considering the effort)
- laggy
- no dynamic model
- does not estimate any higher level variables such as speed - 33next position with mean and median filterMean: c/d
median: D
- 34What is the Kalman approach?The idea whole idea:
We know some things of the dynamics the system and so we predict the next step, then we take our next measurement and compare it to our prediction and adjust our prediction accordingly the next time. - 35What do you need to make a prediction with Kalman filter?Assumption about the behaviour of the system
- 36What is the Kalman gain?relative weight given to the measurements and current state estimate, and can be "tuned" to achieve a particular performance. With a high gain, the filter places more weight on the most recent measurements, and thus follows them more responsively.
- 37Pro's and cons with KalmanPro:
+ dynamic model of the system
+ no lag
+ tunable trade-off between model and measurement
+ it makes an uncertainty estimate
+ cheap to run (everything you need to keep in memory is some matrixes)
Cons:
- parameters not intuitive
- overshooting - 38When should use the Kalman filter?Good for things which don’t have any sharp movements such as planes and ships.
- 39What are four important features of the Kalman filter?Bayesian—It uses Bayes rule to estimate the probability distribution
of the state variables from noisy measurements
Gaussian—All probability distributions in the Kalman filter are
Gaussians, including measurement noise, process noise, and the
state estimates.
Linear— the dynamics of the system being measured are linear.
Online—thee Kalman filter can update the state estimate as soon as
the latest measurement is available. It does not have to wait until all
the data are available, like a batch filter would - 40What if the noise is not gaussian distributed and the system is non-linear?You can use a particle filter
- 41Pros/cons particle filter?+ general
+ continuous or discrete variables
+ great results
- lots of memory
- very slow - 42How can we reduce lag in mean filters?One way to reduce the lag is to
use a weighted average whose weights decrease in value for increasingly
older measurements - 43What is the difference between the Kalman and the mean/median filters?The major difference between the Kalman filter and the mean and median
filters is that the Kalman filter has a dynamic model of the system to keep up
with changes over time. Another advantage of the Kalman filter is that it gives an estimate of its
own uncertainty - 44Why is it useful to have a knowledge of uncertainty in ubicomp?For instance, a Kalman filter
tracker might indicate that a person is equally likely in the kitchen or the living
room, which could affect whether certain automatic actions are triggered. - 45What is one of the main limitations of the Kalman filter?It is the linearity of the
dynamic model. Some processes are inherently linear, such as the radar
traces of ballistic missiles to which Kalman filter were applied a long time
ago. System dynamics are often not linear. For instance, a ball bouncing
off a wall could not be represented by the state system matrix - 46What is the Particle filter?The particle filter is a more general version of the Kalman filter, with less
restrictive assumptions and, because of that, more computational demand.
Unlike the Kalman filter, the particle filter does not require a linear model
for the process in question, and it does not assume Gaussian noise
In terms of the list of properties given for the Kalman filter, the particle
filter is Bayesian, non-Gaussian, nonlinear, and online - 47What is the main advantage of the Particle filter?One of
the main advantages of the particle filter for ubicomp applications is that it
can easily represent an arbitrary mix of continuous and discrete variables
along with a rich model of how these variables interact and affect sensors. - 48The arrival of the internet is considered the "first wave". What is the second?Mobile computing
- 49The Arduino kit is good for..rapid prototyping, connect different sensors, cheaply develop and deploy
- 50What is the best definition of a sensor?Translates a physical magnitude into information
- 51Which of the following is not a sensor? camera, GPS, Calendar, ProjectorProjector
A calendar translates human text entry into information events - 52Which sensor is used for "Dead reckoning" when GPS is unavailable?Inertial measurement unit (f accelerometers, gyroscopes..)
- 53A high signal to noise ratio means..Strong signal among little noise
- 54With a high Kalman Gain, a filter places more weight on..The most recent measurements
Increasing gain -> more weight on more recent measurements - 55When would you decrease the gain of a Kalman filter?When you want the system to respond more slowly to physical events. Such as when you have an automatic door. If you put a really high gain the door is going to open very quickly. But if you want the system to not be so quick you reduce the important of recent value (reduce gain).
- 56What is the purpose of the Kalman gain?To determine how quickly responds to context
- 57Why/When would we use a Kalman filter?You can use a Kalman filter in any place where you have uncertain information about some dynamic system, and you can make an educated guess about what the system is going to do next
Information
- Description
- Mobile comp Flashcards on L3: Sensors, created by Emma Tysk on 03/11/2019.
- Created by
- Emma Tysk
- Views
- 1
- Rating