Politecnico di Torino (logo)

Deep Learning-Based Real-Time Detection and Object Tracking on an Autonomous Rover with GPU based embedded device

Aldo Calio'

Deep Learning-Based Real-Time Detection and Object Tracking on an Autonomous Rover with GPU based embedded device.

Rel. Paolo Garza. Politecnico di Torino, Corso di laurea magistrale in Ingegneria Informatica (Computer Engineering), 2021

PDF (Tesi_di_laurea) - Tesi
Licenza: Creative Commons Attribution Non-commercial No Derivatives.

Download (2MB) | Preview

In recent years, there has been a growing interest in autonomous systems with artificial intelligence, especially in the military, home automation and smart cities sectors. In this project, the aim is to develop, through an artificial neural network capable of detecting objects, a system that can make a rover-type vehicle independent. An autonomous vehicle is a system capable of perceiving its environment and moving safely with little or no human input. Based on an appropriate network according to the needs, a tracking algorithm is developed, which gives the memory to the intelligence, i.e. it can assign an identity to any detected object, around frames in a real-time video. The implementation of a tracking algorithm would allow the system to be more dynamic and to perform more complex functions than simply detecting and tracking the object frame by frame. This paper explores the possibility to create an autonomous system that allows the user to select the desired mode of use by means of a graphic user interface. A Follow Me module is implemented, a technology that is now present in many systems of this type, such as drones or domestic robots, which allows the drone to follow a moving object around autonomously. The Follow Me module allows any person detected to voluntarily activate and, vice versa, deactivate the tracking function of the system. Among the functions implemented in the security module, the identity switch recognizer plays a central role. In the context of multiple objects tracking, one of the most common situations is in fact the overlapping of the target object with other objects in the image. It is therefore necessary to have a security function that is aware of this behavior and warns the system of a possible ID. Computer vision systems have difficulty in maintaining the identity of the individuals over long periods of time; many modern vision systems multiple objects are based on tracking identification, which means they propagate the identities along the track. The following work is part of a larger research project, entitled "INTEGRATION OF OBJECT DETECTION IN REAL-TIME IMAGES IN AUTONOMOUS VEHICLE DRIVING", supported by the National University of Cordoba, aimed at the autonomous operation of a rover-type land vehicle. The tests that will be carried out will relate to the use of a very specific board, the INVIDIA Jetson-Nano, which is mounted in the physical system supplied to the general project.

Relators: Paolo Garza
Academic year: 2020/21
Publication type: Electronic
Number of Pages: 81
Corso di laurea: Corso di laurea magistrale in Ingegneria Informatica (Computer Engineering)
Classe di laurea: New organization > Master science > LM-32 - COMPUTER SYSTEMS ENGINEERING
Ente in cotutela: Universidad Nacional de Cordoba (ARGENTINA)
Aziende collaboratrici: UNSPECIFIED
URI: http://webthesis.biblio.polito.it/id/eprint/18081
Modify record (reserved for operators) Modify record (reserved for operators)