ETD system

Electronic theses and dissertations repository

 

Tesi etd-06282016-194213


Thesis type
Tesi di laurea magistrale
Author
BERTOLUCCI, MATTEO
URN
etd-06282016-194213
Title
Probabilistic Data Association Between Sparse Polygonal Grid Maps And Dynamic Object Representation
Struttura
INGEGNERIA DELL'INFORMAZIONE
Corso di studi
INGEGNERIA ROBOTICA E DELL'AUTOMAZIONE
Commissione
relatore Prof. Caiti, Andrea
controrelatore Prof. Pratelli, Maurizio
correlatore Prof. Pollini, Lorenzo
tutor Loehlein, Otto
tutor Duraisamy, Bharanidhar
Parole chiave
  • sensor fusion
  • data association
  • multi target tracking
  • multi hypothesis tracking
  • polar grid map
  • slow moving vehicles
  • dynamic targets
  • Kalman filter
  • autonomous driving
  • static targets
Data inizio appello
21/07/2016;
Consultabilità
parziale
Data di rilascio
21/07/2019
Riassunto analitico
An important requirement for autonomous driving, is to detect correctly static targets and dynamic targets under various state of motion.<br>The opportunity of detecting that depends upon the availability and reliability of different sensor data processed by the sensor fusion module. This thesis uses data from laser, radar and stereo-camera with built-in tracking modules and the objective is to make the resultant of two different sensor fusion modules, sharing the same sensor data, to be consistent accordingly to their operational requirements. One sensor fusion module deals with dynamic targets providing an L-shape representation, while the other module is built to represent only static targets by means of contour polylines. Due to uncertainty in detecting static targets, a clear differentiation between standing objects and vehicles moving at slow speed is not possible for the second module. Slow moving objects can then be considered as standing objects and wrongly represented by the perception system. The aim of this thesis is then to provide a better representation of vehicles moving at slow speed. This is done, at first, adapting the contour representation for moving objects. Then, data association techniques between the two modules are investigated and a novel approach based on multi-hypothesis tracking is presented. The results are evaluated using simulations and validated with real world data and ground truth.
File