Technical details

The usage of drones in civil applications depends on whether they offer a good level of autonomy for flying and for their mission. The following projects intend to offer that level of autonomy in different scenarios.

Navigating autonomously in indoor and outdoor environments

To do so, the drone ia able to recognize its environment in 3D using only one camera. Currently, we have worked on the following challenges:

• Following a red line located on the floor
• Recognizing signs: the drone can analyze signs so as to take decisions at important points, e.g., turning left/right at a crossing.

• Interpreting the environment in 3D so as to avoid obstacles, and also to minimize landmarks (e.g., avoid putting a red line along all the possible drone paths). To do so, we use two approaches: Dense 3D scanning, and sparse 3D scanning. A dense 3D scan is more precise but forces the drone to stop while scanning. In sparse3D one, the 3D scan is less precise, but the drone can continue flying (cork-screw flight).

Tracking persons and groups of persons

For assisting disasters, identifying and following groups of persons is particularly important in order to predict the necessary assistance. The first video illustrates a tracking in various environments: inside, outside, low light conditions, strong light conditions, light or strong variations in directions. Once again, the drone totally navigates autonomously.

Following a person can also be performed by using a combination of HOG transforms and particle filters  applied on drone images.

Door identification

Doors are quite noticeable elements in buildings. Indeed, they are commonly related to one given room, or one given person. Thus, drone autonomous indoor navigation shall take doors into account. The following video demonstrates how the drone can identify doors and then how it can fly towards them.