âšī¸ââī¸IMU activity recognition
Last updated
Last updated
In this walkthrough, you will learn how to use the Coretex platform for building a model capable of analyzing different stages of a basketball player's jump based on a single BioMech's IMU sensor attached to their lower back.
Equipped with this workflow, you can use it to quickly, reliably and cost-effectively create and commercialize products such as BioMech Putt and BioMech Lab.
After logging in it will bring you to the Project screen with two options, creating new or selecting existing project. Click on the "+ Create new".
A pop-up modal with the title "Create new project" will appear.
Now you need to write all the necessary information for one Coretex Project. In this example, we are training a model that is detecting stages of a correct jump, so for our Project Type we will select the "Motion Recognition" task.
After creating Coretex Project, we are now ready to create our first Task. To create our first task on the left sidebar click on "Workflow". Switch to the 'Tasks' screen using the toggle button.
On Tasks Controls click on "+ New Task". A pop-up modal with the title "Create a new task" will appear. Now you need to write all the necessary information for your Coretex Task.
Select Task Type to "From local folder". Select your script and click upload.
After a few moments, our task is created.
After opening the project on the mobile app go to the dataset screen by clicking on the left-side menu.
On dataset controls click on "Create new Dataset". A pop-up modal with the title "Create a new dataset" will appear. Now we need to write all the necessary information for our new Coretex Dataset.
After entering all the necessary information click on the save button. Automatically after creation, the dataset will be opened, and a button for adding samples is displayed.
Click on "Add Sample". A pop-up modal with the title "Add Sample" will appear. Now we need to write the sample name and click on the Next button.
The interface for recording samples is displayed. Click on enable Video button. With this action, you are allowing Coretex to get access to a camera which will help you with the annotation process later.
Click on "Assign sensor". This will open the panel with all available sensors for recording.
BioMech sensors are used in this case for collecting data.
After the sensor is connected successfully, the button for recording is active. Now you can start with collecting data for a detector.
In this case, we recorded 10 samples which will be used for creating a detector.
We are back on the Coretex Web app. Clicking on Workflow in the sidebar menu will open the Workflow screen. Click on the '+ Add New Workflow' button and a pop-up modal titled "Create a new workflow" will appear.
Click on the 'Insert Task' button and modal with task we previously created will be displayed.
Since training a model is now a work of art, there is no golden rule that we need to follow to get an excellent detector. In this example, we will enter several parameters and run multiple runs at once.
We used the option to run several runs at once in Coretex and set several parameters for vectors, epochs, eventDistance, horizon, noEvents, and batchSize. In this case, we ran 50 runs.
After completing all 50 runs, we chose the run with the highest accuracy, in our case, it is 96%. The live test can be done via the mobile app.
Live testing of the model will be done through the Coretex Mobile app.
Open the "Models" screen through the sidebar. Swiping from right to left will display an option for the Live test.
Click on the "Live Test" button. A pop-up modal with the title "Run Model" will appear. After loading is finished click on the "Next" button.
The interface for testing is opened. As you can notice the interface is similar to the one that we used for recording samples. Assign BioMech sensor. After the sensor is assigned, we are ready to start live testing of a model.
Click on the "Record button" button. Live testing will start. You can start performing movement to see if the model detects events used for training.
On the chart, you will see live data displayed with markers on a position where movement is detected. In the side panel under Classes, you will see how many times one type of class is detected.
Putt Detector is created using Coretex.ai.
Through the process of collecting data in the Coretex.ai mobile app, annotating these same data, and starting runs on the web app we got the best detector that is currently implemented in the product.
Gait Detector is created using Coretex.ai.
We used Coretex.ai mobile app together with the BioMech sensor to record a couple of samples. The same samples are used for the training model. The best model is deployed and used as the current BioMech Lab detector for Gait Toolkit.
