Simone Fassio
Real-Time Deep Learning-Based 3D LiDAR Detection for Multi-Object Tracking in People Flow Monitoring.
Rel. Paolo Garza, Jean-Luc Dugelay. Politecnico di Torino, Corso di laurea magistrale in Data Science And Engineering, 2026
Abstract
This thesis investigates whether learned 3D LiDAR detection can improve person detection for multi-object tracking in large airport terminals while preserving Outsight’s production tracker and meeting a mean 100 ms/10 Hz observation-generation budget. Unlike automotive on-board perception, airport People Flow Monitoring (PFM) processes fused data from many fixed sensors across large areas. The existing geometric stage is fast, but fixed rules can fragment one person, merge nearby people, or discard people with sparse or atypical geometry. The objective is not merely better frame-level detection, but more reliable passenger trajectories. Starting from PointPillars–CenterPoint, this work designs compact bird’s-eye-view networks with sensor-independent features and cylindrical person targets.
Its principal contribution, the Statistical Pillar Feature Encoder (StatPFN), represents each pillar through count, centroid, covariance, and height-distribution statistics instead of point-wise processing
Relatori
Anno Accademico
Tipo di pubblicazione
Numero di pagine
Informazioni aggiuntive
Corso di laurea
Classe di laurea
Ente in cotutela
Aziende collaboratrici
URI
![]() |
Modifica (riservato agli operatori) |
