polito.it
Politecnico di Torino (logo)

Multiple Object Tracking and trajectory prediction for safety enhancement of autonomous driving

Francesca Pacella

Multiple Object Tracking and trajectory prediction for safety enhancement of autonomous driving.

Rel. Enrico Magli, Daniele Brevi, Edoardo Bonetto. Politecnico di Torino, Corso di laurea magistrale in Ict For Smart Societies (Ict Per La Società Del Futuro), 2019

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

Download (20MB) | Preview
Abstract:

In the autonomous vehicles scenario, the V2I communication between vehicles and the road infrastructure is becoming a useful way to increase safety conditions of vulnerable road users. By equipping the infrastructure with a camera, it is possible to handle dangerous situations such as occlusions in a cross road. This can be done by running an object tracking and trajectory prediction algorithm for each road user and inform incoming intelligent vehicles about their position. In this thesis, has been developed a multi-thread framework able to manage different objects tracked at the same time. The framework aims to define the best object tracking algorithm provided by OpenCV C++ library and to solve the Multiple Object Tracking (MOT) and trajectory prediction problem in order to assist intelligent vehicles in their maneuvers. The provided solution is an Online and Partially Detection-Based Multiple Object Tracking algorithm. It runs an online Single Object Tracking algorithm for each single target in the scene. Moreover, it receive information about an accurate detection which runs at lower speed and match each tracker target with its corresponding detection by solving an assignment problem. This allows to reduce tracking drift, improve the online algorithm update and to manage incoming and outgoing objects. The matching algorithm works based on values of extracted feature both on the detection side and on the tracking one. These features are geometrical parameters, color distribution and image key points. Thus, the framework acts as a bridge between the detection stage and the tracking one and as a supervisor for each single tracker. For the algorithms evaluation, a set of KPIs has been defined. Through this evaluation stage, the best OpenCV tracking algorithm has been identified for the specific use case scenario. The framework is able to process video at high frame rate, since it uses easy and lightweight algorithm for the tracking task and it is not dependent on the detection stage. Moreover, thanks to the match stage with accurate detection, the algorithm does not suffer tracker drift problems and also the total number of target to track is continuously kept updated.

Relatori: Enrico Magli, Daniele Brevi, Edoardo Bonetto
Anno accademico: 2019/20
Tipo di pubblicazione: Elettronica
Numero di pagine: 72
Soggetti:
Corso di laurea: Corso di laurea magistrale in Ict For Smart Societies (Ict Per La Società Del Futuro)
Classe di laurea: Nuovo ordinamento > Laurea magistrale > LM-27 - INGEGNERIA DELLE TELECOMUNICAZIONI
Aziende collaboratrici: FONDAZIONE LINKS
URI: http://webthesis.biblio.polito.it/id/eprint/18684
Modifica (riservato agli operatori) Modifica (riservato agli operatori)