Abstract
While motion sensors can detect when a user is nearby, the sparse measurements from sensors within a physical space cannot capture the user’s trajectory. This disclosure describes the use of machine learning techniques to construct the user’s two-dimensional (2D) trajectory within a physical space based on user-permitted motion sensor data by customizing the encoder/decoder embedding layers in the transformer architecture. The encoder maps the sparse input data from individual sensors to a continuous time sequence of mapped sensor events that are provided to the decoder. The decoder can couple the output of the encoder with a 2D array describing the user’s trajectory to generate predicted 2D coordinates for the user’s next position in the trajectory. Since the transformer architecture is autoregressive, the 2D array representing the user’s trajectory is updated with the most recent estimate of the user’s position to serve as the next input to the decoder. The constructed trajectory can be used to facilitate intelligent operations and services in the physical space.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Shin, D, "Constructing Motion Trajectory from Sparse Readings from Distributed Sensors", Technical Disclosure Commons, (February 09, 2024)
https://www.tdcommons.org/dpubs_series/6673