Cristina Nuzzi
As an Industrial Engineer I think automation is the future of industries. An high number of processes can be automated today, thanks to the joint expertise and knowledge of the mechanical and industrial world and the computer science world, which is usually more consumer oriented. That's why over the years I've tried to acquire knowledge from both worlds, and use this knowledge to research more industry-oriented applications of algorithms and systems, such as Deep Learning algorithms for Machine Vision Systems, and test their capabilities from a measurement and instrumentation point of view.
Supervisors: Giovanna Sansoni
Supervisors: Giovanna Sansoni
less
Related Authors
InterestsView All (9)
Uploads
Papers by Cristina Nuzzi
We acquired four different small datasets with different characteristics to evaluate the performances in different situations: a first dataset in which the actors are dressed in casual wear; a dataset in which the actors wear skin-like color clothes; a dataset in which the actors wear light-blue gloves; and a dataset in which the camera is placed close to the operator. We tested the performances of the model in two experiments: first by using a test dataset composed of images of actors who were already present in the corresponding training dataset and second, by using a test dataset composed only of images of a chosen operator not present in the corresponding training dataset.
Our experiments show that the best accuracy and Fl-Score are achieved by the Complete dataset in both cases, and that the performances of the two experiments are comparable. We tested the system in real-time, achieving good performances that can lead to real-time human-robot interaction with a low inference time.
We acquired four different small datasets with different characteristics to evaluate the performances in different situations: a first dataset in which the actors are dressed in casual wear; a dataset in which the actors wear skin-like color clothes; a dataset in which the actors wear light-blue gloves; and a dataset in which the camera is placed close to the operator. We tested the performances of the model in two experiments: first by using a test dataset composed of images of actors who were already present in the corresponding training dataset and second, by using a test dataset composed only of images of a chosen operator not present in the corresponding training dataset.
Our experiments show that the best accuracy and Fl-Score are achieved by the Complete dataset in both cases, and that the performances of the two experiments are comparable. We tested the system in real-time, achieving good performances that can lead to real-time human-robot interaction with a low inference time.