On Android | Edge Impulse Documentation

Home API & SDKs Forum Studio

Impulses deployed as a, can be included in an Android application to run the impulse as an Android distributable binary (APK). This tutorial will demonstrate some examples of building an Android application around image or sensor data.

This is made possible by using the, which allows you to write native code in C/C++ and include it in your Android application. The C++ library generated by Edge Impulse can be included in your Android application to run the impulse on device.

Try the example above on your Android device:

Prerequisites

Also install the following software:

Clone the base repository

We created an example repository which contains a sample application for Android, and wearOS which you can use to build on, and experiment using your own impulse. Download the application as a .zip, or import this repository using Git:

git clone https://.com/edgeimpulse/example-android-inferencing

Deploy your C++ project

Import your C++ project

Depending on the example you want to use, import the project into Android Studio:

Static buffer

The Static Buffer example is a simple application that uses a static buffer to run the impulse on the device. The application will show the result of the inference on the screen.

To get inference to work, we need to add raw data from one of our samples to native-lib.cpp. Head back to the studio and click on Live classification. Then load a validation sample, and click on a row under 'Detailed result'. Make a note of the classification results, as we want our local application to produce the same numbers from inference.

Here we replace the raw_features array in native-lib.cpp with the raw data from the sample.

The application will show the result of the inference on the screen.

Android

The Android example is a simple application that uses the camera to collect data, and run the impulse on the device. The application will show the result of the inference on the screen.

WearOS

The WearOS example is a simple application that uses the accelerometer sensor to collect data, and run the impulse on the device. The application will show the result of the inference on the screen.

Add Edge Impulse C++ files

Unzip your Edge Impulse C++ Library export. Copy these folders into your project's app/src/main/cpp/ directory:

edge-impulse-sdk/
model-parameters/
tflite-model/

Do not copy the CMakeList.txt file.

The CMakeList.txt that is included in the repo was modified in the for Android applications.

Download the TFLite libraries

Run the Windows / Linux / OSX script to fetch resources

cd example-android-inferencing/example_static_buffer/app/src/main/cpp/tflite
sh download_tflite_libs.bat # download_tflite_libs.sh for OSX and Linux

Now you can build the application.

Building the application

To build the application, open the project in Android Studio, and click on the 'Run' button. This will build the application and deploy it to your Android device.

Adding additional sensors

If you want to integrate additional sensors, such as a Gyroscope or Heart Rate Sensor, follow these steps:

Enable the Sensor in the Code In MainActivity.kt, locate the sensor initialization section and uncomment the corresponding lines:

// Uncomment to add Gyroscope support
private var gyroscope: Sensor? = null

// Uncomment to add Heart Rate sensor support
private var heartRateSensor: Sensor? = null

Initialize the Sensor in onCreate Inside onCreate(), uncomment and initialize the sensor:

 gyroscope = sensorManager.getDefaultSensor(Sensor.TYPE_GYROSCOPE)
// heartRateSensor = sensorManager.getDefaultSensor(Sensor.TYPE_HEART_RATE)

Register the Sensor in onResume To start collecting sensor data when the app is active, uncomment the registration logic:

 gyroscope?.also {
     sensorManager.registerListener(this, it, SensorManager.SENSOR_DELAY_NORMAL)
 }

// heartRateSensor?.also {
//     sensorManager.registerListener(this, it, SensorManager.SENSOR_DELAY_NORMAL)
// }

Handle Sensor Data in onSensorChanged Modify the onSensorChanged() function to collect new sensor data:


 Gyroscope data
 Sensor.TYPE_GYROSCOPE -> {
     ringBuffer[ringBufferIndex++] = event.values[0] // X rotation
     ringBuffer[ringBufferIndex++] = event.values[1] // Y rotation
     ringBuffer[ringBufferIndex++] = event.values[2] // Z rotation
 }

// Heart Rate data
// Sensor.TYPE_HEART_RATE -> {
//     ringBuffer[ringBufferIndex++] = event.values[0] // Heart rate BPM
// }

Unregister the Sensor in onPause To save battery and improve performance, ensure sensors stop when the app is paused:


sensorManager.unregisterListener(this)

Update the CMakeLists.txt for additional libraries or options (optional step for entire projects e.g. HRV)

This is the where you can add additional libraries or options to the CMakeLists.txt file. For example, to enable the full TFLite library, you can add the following line to the CMakeLists.txt file:

add_definitions(-DEI_CLASSIFIER_ENABLE_DETECTION_POSTPROCESS_OP=1
    -DEI_DSP_ENABLE_RUNTIME_HR == 1
    -DEI_CLASSIFIER_USE_FULL_TFLITE=1
    -DNDEBUG
)

Hardware Acceleration (Coming soon)

To further optimize inference on Android, future updates will include:

GPU acceleration with LiteRT delegate: Improves performance for TFLite models.

Conclusion

You now have a working Android application that runs your impulse on the device. You can use this as a starting point to build your own application, or integrate it into an existing application.

Android opens up the ability to distribute Android APKs for a wide range of platforms, including WearOS, Automotive, Television, Unity, and eXtended Reality (XR). This ease of distribution is a powerful tool for deploying your impulse on a wide range of devices.