| KEYWORD |
Hybrid multi-LidAR sensor fusion and objects detection and recognition
keywords LIDAR, OBJECTS DETECTION, RECOGNITION
Reference persons DANIELA ANNA MISUL
Research Groups PT-ERC
Thesis type SIMULAZIONE NUMERICA
Description LiDARs are mainly used for perception and localization. The output of a perception system comprises three levels of information: 1) Physical description (velocity, shape etc)2)Semantic description (categories of objects) 3)Intention prediction (likelihood of an object’s behavior).Therefore, the LiDAR outputs are used for the object detection, classification, tracking, and intention prediction.Usually LiDARs are combined with cameras to complement each other. For an autonomous vehicle, its perception system classifies the perceived environment. A traditional workflow consists of four steps: object detection, recognition, tracking, and motion prediction. Object detection algorithms extract the objects’ physical information (position and shape). Object detection comprises ground filtering (meant to label a point cloud as ground/non-ground) and clustering (meant to group nonground points into different objects). Object recognition algorithms provide the objects’ semantic information (pedestrian, vehicle, building, etc). What has been done so far:
Introduction of solid-state LiDAR within APOLLO environment.
Segmentation CNN employed for LiDAR objects detection & recognition
Thesis objectives:
Hybrid multi-LiDAR sensor fusion; Implementation of a neural network architecture in APOLLO with satisfactory performance for hybrid multi-LidAR signals
Required skills python; C++
Deadline 13/03/2023
PROPONI LA TUA CANDIDATURA