This is the Python implementation for the Paper: IMUNet

A new Architecture called IMUNet which is appropriate for edge-device implementation has been proposed for processing IMU measurements and performing inertial navigation.

In this repository, the data-driven method for inertial navigation proposed in here for the ResNet18 model has been modified. Other than ResNet18, four state-of-the-art CNN models that have been designed for IoT device implementation have been reimplemented for inertial navigation and IMU sensor processing.

Other than IMUNet, MobileNet, MobileNetV2, MnasNet, and EfficientNetB0 models have been re-implemented to work with one-dimensional IMU mesurements.

Five datasets have been used in the paper.

• A new method for collecting a dataset using Android cellphones that uses ARCore API for collecting the ground truth trajectory has been proposed and a dataset using this method along with the method proposed in RIDI using a Lenovo Tango device for collecting the ground truth trajectory has been collected. A preprocessing step has been added to read and prepare the data. The collected dataset can be downloaded from IMUNet_dataset.

Other datasets are:

1- RONIN which is available at here

2- RIDI which is available at DropBox

3- OxIOD: The Dataset for Deep Inertial Odometry which is available at OxIOD

4- Px4 which can be downloaded from px4 and the scripts provided here has been used to download the data and pre-process it.

The data-driven method for inertial navigation proposed in RONIN for the ResNet18 model with all the new architectures and datasets as well as the proposed architecture have been implemented in Tensorflow-Keras.

The Android Application is available at Android. It contains three sections:

1- The application for collecting a new dataset.

2- The test part of the RONIN for ResNet18 model using all the proposed models and some samples of the collected dataset.

3- A comparison has been implemented to show the efficiency and accuracy of the proposed model. The result can be seen in the video below: