Indoor Navigation project: recognition of human movement paths based on signals from IMU sensors of wearable devices
Project with Sensor Team Huawei

The project team was assigned the task of restoring a two-dimensional trajectory of human movement in space based on the signals of IMU sensors of a mobile device: accelerometer and gyroscope. It was necessary to implement deep learning models to solve the problem with a high degree of accuracy. The customer provided marked-up data and specified requirements for the accuracy of the algorithm.


Wearable devices must recognize the trajectory of movement in situations where there is no GPS signal and it is impossible to track the trajectory using it. Such situations occur both inside buildings and in dense urban development.
The task of the project to create neural network models is relevant because of the limitations of classical methods of Indoor Navigation in the speed and accuracy of forecasting.
The decision of the team
A full cycle of research was conducted on the methodology for implementing research projects created by the laboratory team. The project included research of the literature and patent base, as well as building a system for testing solutions, as well as selecting the most accurate models. As a result, neural network models of trajectory recovery were built, which are based on recurrent neural networks. During the research work, two articles were written and a patent was issued for Huawei.
- Customer-patented solution
- Low recognition error
- 2 scientific articles
Permitted difficulties
- Determination of the free fall vector
- Trajectory drift
- Building an ODE model
- Project Manager: Alexey Goncharov
- Team Lead: Ilya Zharikov
- Team of researchers: Philip Nikitin, Tamaz Gadaev
- Scientific consultant of the project: Vadim Strizhov
Technology stack
Python, PyTorch, TensorFlow
Made on