DSTN Scholars

DSTN > Scholars > OLOU Babatoundé Hervé

Title: Design and integration of deep learning for adaptive and contextual control of unmanned aerial vehicles: application to environmental monitoring 

Affiliated project Deep4EnvMonitoring

Affiliated CEAUniversité d'Abomey-Calavi: CEA SMIA & Université Gaston Berger: CEA MITIC

Thesis director : Eugène EZIN, Institute of Mathematics and Physical Sciences, University of Abomey-Calavi, CEA SMIA, Benin, eugene.ezin@gmail.com

Thesis co-supervisor : Jean Marie DEMBELE, UFR SAT, CEA MITIC, Senegal, jmdembele@gmail.com

Other contributors to thesis supervision :
Christophe CAMBIER, Sorbonne-Université - IRD
Cédric HERPSON, Sorbonne-Université, - LIP6

Scientific publications:

[1] Olou, H. B., Ezin, E. C., Dembele, J. M., & Cambier, C. (2022, November). FCPNet: A novel model to predict forward collision based-upon CNN. In 2022 22nd International Conference on Control, Automation and Systems (ICCAS) (pp. 1327-1332). IEEE.
[2] Olou, H. B., Ezin, E. C., Dembele, J. M., & Cambier, C. (2022, November). Forward Obstacle Detection by Unmanned Aerial Vehicles. In Pan-African Artificial Intelligence and Smart Systems Conference (pp. 397-410). Cham: Springer Nature Switzerland.
[3] Olou, H. B., Ezin, E. C., Dembele, J. M., & Cambier, C. (2023). Global Motion Planning for Unmanned Aerial Vehicle Automation. In 2023, 6th International Conference on Intelligent Autonomous Systems (ICoIAS). IEEE.
[4] Olou, H. B., Ezin, E. C., Dembele, J. M., & Cambier, C. (2023). Autonomous Navigation of Unmanned Aerial Vehicle: Investigating Architectures and Techniques for a Flexible Platform. Unmanned Systems, 1-17.

DSTN > Scholars > OLOU Babatoundé Hervé

Start date: 01/01/2021
Anticipated date of thesis defense: May 2024
ORCID profile: 0000-0002-2607-2809

Unmanned aerial vehicles (UAVs) are flying robots that are currently widely used in many fields, both military and civilian. Their development is an area of research in automation and robotics, attracting the interest of a vast community. UAVs are particularly useful in environmental monitoring, contributing to waste detection, land surveillance or precision agriculture, and facilitating various analysis and control missions. The aim of this work is to implement an autonomous drone using artificial intelligence techniques, and to generate optimal trajectories for autonomous localization and navigation. This work aims to develop an artificial intelligence capable of adjusting the flight plan to modulate the overflight altitude according to the specific context, as well as the definition and frequency of captures, integrating if necessary the use of other terrestrial sensors.

Objectives / Expected results :

  • A flexible platform is proposed in [4], which can be easily followed to implement an autonomous UAV according to the mission.
  • A new model is built to predict head-on collisions based on CNNs. This model is used on a Raspberry pi 4 and shows excellent results [1].
  • After collision prediction, an algorithm is proposed in [2] to detect the position of obstacles and indicate the next direction to the UAV.
  • During the course of this thesis, no open-access tool was found to help us define a no-fly zone and autonomously generate a path to avoid these zones. To this end, in [3], we modified the QGroundControl source code and integrated a no-fly zone definition module and an autonomous path generation module. Furthermore, in [3], an algorithm is proposed to regenerate the trajectory during the mission when an obstacle is detected.


Prospects for the end of the thesis:

  • Building detection using segmentation techniques and satellite images

Prospects after completion of thesis:

  • Collect new data and retrain models for better performance
  • Purchase new equipment to continue autonomous drone deployment
  • Continue to set up a web-based supervision platform
  • Develop a new module in QGroundControl for automatic detection of hazardous areas using satellite images.

Contribution / added value to the affiliated project :
In this program, there is another thesis on evolutionary hyper network for convolutional neural networks (CNN). In collaboration with this thesis, the present thesis will adopt and apply the generalization of the evolutionary hyper network to UAV control in order to address different environmental monitoring problems.