Tizen Native (Wearable) NNStreamer Application Example - Orientation Detection

Introduction

This example passes accelerometer sensor data stream to a neural network (tensorflow-lite) via tensor_src_tizensensor gstreamer element. The neural network predicts one of four orientation of the device:

1. 12 o'clock is upward.
2. 3 o'clock is upward.
3. 6 o'clock is upward.
4. 9 o'clock is upward.

Description

• This is a sample application of Tizen Native for wearable device.
• If you want to run it on your device, Tizen 6.0 or higher is required.
• About details of NNStreamer, please check this page.
• Used gstreamer pipeline:
  tensor_src_tizensensor (accelerometer) -- tensor_filter -- tensor_sink
  • The accelerometer measures the device's accelerometer vector in 3 axes.
  • The tensor_src_tizensensor element feeds those three float values into tensor_filter element.
  • TF-lite model orientation_detection.tflite predicts possibility of each four orientations.
  • tensor_filter element (with the TF-lite model) provides the tensor with four float values (the possibilities) into tensor_sink

Screenshot