Name: RANIK GUIDOLINI
Publication date: 22/10/2024
Examining board:
Name |
Role |
|---|---|
| ALBERTO FERREIRA DE SOUZA | Presidente |
| ANSELMO FRIZERA NETO | Examinador Interno |
| CLAUDINE SANTOS BADUE | Examinador Interno |
| KARIN SATIE KOMATI | Examinador Externo |
| MARIELLA BERGER ANDRADE | Examinador Externo |
Summary: Most self-driving cars adopt Occupancy Grid Maps (OGM) to navigate instead of Seman-
tic Maps or Topological Maps because OGMs are dynamic, flexible and can be computed in
real time. In OGMs, a grid divides the operational environment into small cells that hold the
probability of each small region of the environment being occupied by an obstacle. Movable
objects can leave traces on the OGMs as they are in motion, and cells in the OGM marked as
occupied due to the presence of these moving objects may not be revisited or sufficiently
scanned by the sensors to be updated as unoccupied, compromise localization and navigation.
The presence of movable objects traces is misleading information for several aspects of self-
driving cars’ operation (i.e., localization), and are generally removed from OGMs manually, a
time-consuming and prone to errors task. In this work, we propose a technique for removing
traces of movable objects from occupancy grid maps based on deep neural networks, dubbed
Enhanced Occupancy Grid Map Generation (E-OGM-G). In E-OGM-G, we capture camera
images synchronized and aligned with LiDAR rays, detect movable objects in these images
using a deep neural network and compute which laser rays of the LiDAR hit objects classified
as movable objects. By clustering laser rays that are close together in a 2D projection, we are
able to identify clusters that belong to movable objects and avoid using them in the process of
generating the OGMs – this allows generating OGMs clean of movable objects traces.
We evaluated our automatic OGM cleaning technique – dubbed enhanced OGM genera-
tion (E-OGM-G) – in two datasets in more than 7.7 km of a road of the city of Vitória-ES,
Brazil, in normal traffic. E-OGM-G achieved precisions of up to 100%. In addition, we tested
our self-driving car, IARA, on this road and it was able to operate autonomously using the clean
OGMs obtained with our technique. These results show that E-OGM-G is effective in removing
moving objects from static OGMs.
