Self-optimizing neural networks for efficient person localization and tracking
keywords EMBEDDED SYSTEMS, INDOOR LOCALIZATION, NEURAL NETWORKS, STRUCTURAL OPTIMIZATION, EVOLUTIVE ALGORITHMS, RE
Reference persons LUCIANO LAVAGNO, MIHAI TEODOR LAZARESCU
Research Groups Sensing and processing, microelettronica
Thesis type RESEARCH
Description The broader project currently involves both Ph.D. and MS students and is aimed at the identification (out of a small set), localization, and tracking of persons indoors, using sensor data fusion from newly developed sensors within the project (e.g., long-range capacitive) and existing ones, operating on established sensing principles (e.g., very low-resolution infrared camera).
Specifically, the topic consists in searching the state-of-the-art for existing neural networks solutions that can either:
- optimize themselves during learning, to achieve an efficient implementation (minimize memory, processing) for inference;
- and/or are capable to tune and make use of “tools”, i.e., hard-coded algorithms (e.g., mathematical functions) in order to minimize the necessary resources (memory, processing) for inference.
The most promising NN solutions can then be implemented and evaluated using the experimental data that we have collected within the broader project for indoor human localization and tracking.
See also https://iot.det.polito.it/indoor-human-localization-and-identification-using-low-power-low-cost-long-distance-capacitive-sensors/
Deadline 04/10/2023 PROPONI LA TUA CANDIDATURA