INRAE

When:
30/11/2021 – 01/12/2021 all-day
2021-11-30T01:00:00+01:00
2021-12-01T01:00:00+01:00

Offre en lien avec l’Action/le Réseau : BigData4Astro/Doctorants

Laboratoire/Entreprise : Romea team, INRAE Clermong-Ferrand
Durée : 6 mois
Contact : zhongkai.zhang@inrae.fr
Date limite de publication : 2021-11-30

Contexte :
The equipment of 3D Lidar sensor for mobile robot navigation
allows to obtain real-time point clouds. Compared with 2D RGB
image, 3D point cloud describes more information of the environment. However, object detection from 3D point cloud is more changing than their 2D counterpart, especially for disordered point cloud. Object detection using supervised learning methods needs a huge annotated data, and when the background changes, new labels should be annotated again. Generative Adversarial Networks (GAN) have been proposed to reduce the annotation task by human for 2D image segmentation. Although GANs have been adapted for 3D point cloud generation, it is not clear how to achieve unsupervised object detection from point cloud using GANs. Therefore, the main objective of this master project is to explore an efficient GAN architecture to detect object of interest from 3D point cloud, and employ it in agricultural fields for the detection of plants, roads and obstacles. Results aims at the representation of agricultural environment and the guidance of off-road mobile robot. The proposed trainee will take part of experiments using a 3D Lidar sensor available at INRAE.

Sujet :
Object Detection in Agricultural Fields using 3D LiDAR Point Cloud

Profil du candidat :
M2 in AI or Robotics

Formation et compétences requises :
M2 in AI or Robotics:

Technical Skills: deep learning, computer vision, robotics
Software: Python, C++, Pytorch, ROS
Language: English

Adresse d’emploi :
Clermont-Ferrand

Document attaché : 202111290755_Object Detection in Agricultural Fields using 3D LiDAR Point Cloud.pdf