User Guide
Achieving exciting outcomes requires that you understand a few simple principles and good practices. We strongly recommend that you read this guide thoroughly before starting.
This technology allows your app to recognize different movements the user is performing and react accordingly.
Some usage examples:
- A worker in a noisy environment raises his arm to signal a crane operator to stop.
- Parents want to check if their children are brushing their teeth.
- One can remind only those people not drinking enough water to start drinking.
- You can turn the lights with a wave of a hand.
… and many more, just let your imagination run wild!
Unlike other technologies that confine you to a specific set of movements, our solution is based on very flexible machine learning, allowing you to identify chosen gestures.
A short overview video is available at: https://youtu.be/LMQi5syT7bU
A more detailed how-to video is available at: https://youtu.be/rR6rM7A8ouo
First, you use our GREAT (Gesture REcording And Testing) app to record a set of repeating movements. The recording is then sent to our cloud service that analyses the recording and sends back an algorithm for identifying the recorded movement. You may use our app also for testing and fine-tuning the algorithm. When you are satisfied with the results, you can download the algorithm and make it part of the app you build.
You are getting the opportunity to experience an emerging technology that is still under development.
As this is not an officially released product, and as improvements are underway, your experience might still not be ideal. However, after some iterations and fine-tuning, you should be able to achieve useful outcomes.
The app contacts a web service to generate the gesture recognition algorithm for your recorded movements. To use it, you need to obtain your free activation code at http://wearables.mybluemix.net/
You may use any device that has an available connector to our framework (currently you may use TI Sensor Tag, Microsoft Band or Gemsense).
Make sure your device is up and running (some devices, such as Microsoft Band, require you to first pair them with the phone using the device manufacturer's app).
To connect to a device, select it from the list and tap Connect.
Tap the '+' button to record your gesture. Fill-in a name and tap Start Recording.
Make sure you are ready to start performing the gesture. There is a 3 second delay before recording begins.
Tip: you must start your movement exactly when recording begins. Otherwise, the system may consider the 'resting' period prior to the gesture as part of the gesture itself, leading to unexpected results.
TIP: A gesture should take up to 3 seconds to complete, ideally 1 to 2 seconds.
A gesture should be clear and distinctive and at the same time natural and ordinary. Avoid highly delicate movements and 'robotic', unnatural movements.
When you've completed the gesture, tap the End button.
Tip: Be ready to tap End exactly when the gesture is completed. Again, being late may lead to unfavorable results. Train yourself a few times or consider asking somebody tap the button for you.
Your gesture will now be validated to make sure it has been captured appropriately.
You will now be asked to perform the gesture again. Each of these repetitions is called an 'Iteration'. To assure satisfactory results, you must record at least 4 Iterations. From 5 to 10 iterations may in some cases yield even better results. Perform subsequent iterations similarly to previous ones.
When you completed all the iterations, tap Finish Recording. Your recording will now be sent to a web service that will analyze it and prepare a corresponding algorithm for you.
If you want to use the algorithm in your application, you should download it using the link shown. Click here here to learn how to make the algorithm part of your app.
Test your gestures to see how they are recognized and to fine-tune some parameters.
Tap Test to start. You may also delete a gesture by left-swiping on it or edit its threshold by tapping and holding the threshold value.
Perform a gesture the way you recorded it.
When a gesture is recognized it is shown at the top of the screen together its likelihood score (which indicates the level of certainty in identifying the gesture). A recognized gesture is shown for a few seconds.
Click advanced if you want to get more visibility and control when you fine-tune your solution.
The table at the bottom of the screen shows all the other gestures the algorithm currently 'suspects', together with some of their parameters:
- Score - an indication to the 'quality' of the gesture performed.
- Threshold - this parameter 'protects' you from false-positives. Even if the likelihood of the recognition is high, the score must be higher than the threshold to assure the recognition is not result of noise.
These gestures are not recognized either because their likelihood score is too low or because their score is lower than the defined threshold.
The default threshold for each gesture is suggested by the algorithm. If you are certain you've performed a gesture and its score appears to be lower than its threshold, you may consider lowering the threshold for that gesture. Be careful as too low a threshold may lead to false-positive recognitions.
Tip: If you are at rest and a gesture is shown with a significant score, it might be because, while recording, you either hesitated before starting the movement or waited after you finished and before you tapped the End button. Consider deleting the gesture and recording it again.