Convolutional Networks for Vehicle Identification on Viaducts
Reference persons DANIELE JAHIER PAGLIARI
External reference persons Alessio Burrello (University of Bologna)
Thesis type EXPERIMENTAL, SOFTWARE DEVELOPMENT
Description During the thesis, the candidate will develop machine and deep learning models for the recognition of vehicles on an Italian viaduct, exploiting the accelerometer data from a pre-existing structural health monitoring installation, which has been synchronized with camera data in order to obtain ground truth vehicle identification labels. The development of the thesis will be divided into three phases:
- Analysis of the models developed in the state of the art. In particular, we will analyze supervised and unsupervised algorithms, to understand if a well performing algorithm can be realized even without the presence of training labels;
- Analysis of the task and its simplification, that is traffic load monitoring.
- Application of advanced automatic neural architecture search (NAS) developed within the research group to improve the performance of the network and reduce its size as well as the energy consumption.
The final outcome of the thesis will be a Machine Learning model, either supervised or not (autoencoder) to solve a novel task with proprietary data coming from the real world. Note that solving this task with accelerometer data would solve several issues of alternatrive solutions, e.g., those based on cameras, improving privacy and reducing the installation costs to 0.
Required skills Required skills include Python programming and a good familiarity with machine/deep learning concepts and models. Basic C knowledge could help in the last phases of the thesis.
Notes Thesis in collaboration with Prof. Luca Beniniís research group at the University of Bologna and ETH Zurich. The thesis can be carried out either in Torino or in one of the other two universities.
Deadline 13/06/2023 PROPONI LA TUA CANDIDATURA