Mirco De Marchi

Mirco De Marchi

Ph.D. student in Computer Science

Verona, Italy

About me

Mirco De Marchi is a Ph.D. student at the forefront of research in real-time sensor fusion and human pose estimation on energy-efficient edge devices. His activity takes place at the Department of Engineering for Innovation Medicine, University of Verona, in Parallel Computing laboratory with supervisor Prof. Nicola Bombieri. His work focuses on the integration of wearable IMU sensors and camera-based sensors to achieve accurate and real-time tracking of human movements on embedded and low-cost devices.

Research Areas

In the field of human pose estimation, multi-person scenarios encounter several difficulties, especially in situations where remote monitoring, patient training, or tracking an operator near a machine necessitate the ability to distinguish the subject being monitored from others. However, when monitoring is carried out using wearable sensors, identification becomes inherent, and by integrating data from multiple sensor sources, the accuracy of recorded movements can be significantly enhanced. Therefore, a methodology is currently being developed to track and identify a specific human body within a multi-person motion capture system, employing wearable wireless IMU sensors. This designed method incorporates motion analysis, utilizing both geometric matching models and neural network-based models to achieve its objectives.

The need for remote motion analysis to gather clinical data on patients' activities or to activate alarm systems is becoming increasingly urgent, particularly for elderly users or in remote and difficult-to-reach areas. In response to this need, a privacy-aware platform has been developed, which allows for 3D human pose estimation at the edge to effectively detect patient falls. Additionally, the platform enables the real-time transmission of clinically relevant data that can be accessed remotely. The project has been funded by the Veneto region and the mountain union of Asiago. Its primary objectives are to detect alarm states in elderly individuals and to enhance the analytical data for the rehabilitation of patients with Parkinson's disease or those recovering from a stroke.

With the imminent rise of Industry 4.0 and human-machine collaboration, human pose estimation systems are playing a crucial role in tasks such as pose forecasting, collision detection with machinery, and identifying operator danger zones in assembly lines. As a result, multi-camera human pose systems are becoming increasingly popular to tackle the occlusion problem and enhance human pose accuracy. To meet these demands, a robust multi-person, multi-view human pose tracking system has been developed, ensuring real-time performance on at-the-edge devices. This distributed system was successfully implemented and tested in a real-world scenario at the I.C.E. laboratory.

Working with resource-constrained, low-cost, and low-energy devices, as well as distributed systems, necessitates the optimization of processes and communication between them. Notably, efforts have been made to optimize convolutional neural networks through pruning and quantization, reducing the computational load of the human pose estimation model. Domain adaptation of the estimated pose with teacher-student architecture at the edge using knowledge distillation and active learning. Additionally, a methodology has been proposed to automate the deployment of nodes in a cluster, enhancing the communication protocol with the ROS interface by utilizing zero-copy techniques. This approach aims to further optimize the overall performance of the system.



Doctoral Program in Computer Science

University of Verona

February 2022

Master's Degree in Computer Engineering for Robotics and Smart Industry

University of Verona

October 2019
October 2021

Bachelor's Degree in Computer Science

University of Verona

October 2016
July 2019


Tutorship grant

Programming, Advanced Computer Architecture, Operating System

2019 2023

Conference Presentation

IEEE Latin American Test Symposium (LATS)



Summer school at Barcelona Supercomputing Center in AI on GPUs


Research grant

ROS library for shared-memory architectures

2021 2022


Optimization of Communication Protocols for Heterogeneous Edge Devices


Research grant

Optimization of a low-cost and low-energy embedded platform

2020 2021

Research grant

Speech recognition models in smartphone for Restaurant Industry

2020 2021


MentOS: an educational Operating System for Memory Management

2018 2019