Cutting-edge AI
Choose the Open Rescue solution that suits your needs
Are you a DJI user? Contact us to integrate it into your existing system.
Seamless integration into the DJI ecosystem
Want to start using Open Rescue right away? Order a Windows or Linux license directly and get up and running in less than 15 minutes. A turnkey solution for efficient search and rescue operations.
Immediate use with your drone
If you're a drone solution developer, you can integrate Open Rescue into your existing services on-premises or on a server instance. Contact us to access our API and integration tools.
Integration into your drone solutions
(for developers)
Potential Uses
Local | Integrated | |
|---|---|---|
Types d'usage | Licence ordinateur | Dans votre serveur |
Use | AI analysis in the field, without connection | Open Rescue AI, integrated into your systems |
Compatible Drones | All drones | Drones or any type of video stream |
Offline | ✔ | ✔ |
Online | - | ✔
(on server) |
Model size IA | N | Custom made |
RGB photo processing | ✔ | ✔ |
Processing on Livestreaming | ✔
(on RCs with an HDMI output) | ✔ |
(Version ?) du modèle | R2n | R2n |
Open Rescue is capable of conducting searches using visible-light imagery. We are currently developing thermal imaging integration, which will complement the RGB data.
Are you interested in this development?
Please feel free to contact us:
You don't need an internet connection! Thanks to its offline mode, it offers flexible and accessible use wherever you are.
Open Rescue is currently being trained on various types of terrain:
-
Mid-mountain
-
High-mountain
-
Plains
-
Cliffs
-
Coastal areas
-
Coniferous and deciduous forests
-
Cliff edges
-
Scree slopes / lapiaz
-
Rivers
-
Waterfalls
-
Urban / semi-urban
-
Lakes / beaches
-
Quarries
-
Glaciers
-
Deserts
The dataset, compiled since 2023, accounts for seasonal variations.
It has also been trained on avalanche zones, cliff rescue scenarios, flooded areas, and accident sites.
Its development by our AI experts, based on 18,000 aerial photos, makes it a valuable tool for your research missions.
This database was created with the participation of many firefighters and extras from the French Hiking Federation of Haute-Savoie, thus offering a diverse representation of the population. The "victims" present in the images of the dataset are, therefore, volunteers pretending to be in danger. No human beings were injured during the data collection.
FAQ
Open Rescue, the AI adapted for you
Together, let's promote your ideas and your needs
Our flexible and responsive team is always listening to your suggestions to improve our solutions. Whether it is to propose improvements, request new features or explore customization options, we work closely with you to develop an AI that perfectly meets your specific requirements. Together, we build ever more powerful tools adapted to your challenges in the field.
Open Rescue, the AI adapted for you
Sensors used for
data capture :
-
Phantom 4 PRO TRK
-
P1 camera on 300 sensor
-
H20 camera on 300 sensor
-
Mavic 3T
-
Mavic 3E
-
M30T
-
Phantom 4M
-
Phantom 4 PRO
-
Mavic 2 ED
-
Phantom 3 PRO
-
Matrice 4E
-
Matrice 4T
Only RGB sensor data was used (no multispectral or thermal data).Note: This list includes the models used to create the dataset. The use of Open Rescue is not limited to these devices; any drone capable of providing standard images is compatible.
The Dataset :
The dataset consists of 26,686 images. After cutting into standard 640 x 640 images for training, this represents 1.35 million images. These cut images are larger or identical in size to datasets such as Imagenet or COCO.
The results are evaluated on a test set representing 10% of the total, in addition to a field validation through several real operations aimed at detecting a single person, as in practical situations.
Once trained, the AI achieved the following performance on its test dataset. In all curves, the term "person" refers to the detection of signs of life associated with people, such as clothing, vehicles, backpacks, individuals. These annotations were made in accordance with the requirements of the pilot SDIS
The following ROC curve is thus obtained.
