The primary objective is data collection for technology development, specifically training computer vision models. This involves defining “scenes” as specific locations with interesting buildings.

For each scene, the requirement:
Five drone flights approaching a designated building from different angles (corners and top-down) and ground-level photos of its four faces taken from a cell phone.

REQUIREMENT GUIDE:
https://flipbooklets.com/pdfflipbooklets/drone-vet-vision-data-capture

We Need a minimum of the following shots per scene:

1. Top down: a. start at max altitude and descend straight down on to the target building. Get as close as reasonably possible.

2. Top down with lateral arc: a. Start at max altitude and then fly in an arc going forward and down towards the target. (Imagine the second half of a thrown ball where it is descending towards
the ground).

3. 45 degree Approach #1: An approach towards target building with approximately 45 degree down angle in the approach.

4. 45 Approach #2: A second approach towards target building with approximately 45 degree down angle in the approach, coming from a different direction as shot #3.

5. Ground level: cell phone photos of target building and any structures within center frame of we are flying towards (shot from all sides):
     5.a We want 360 degree coverage of the buildings
     5.b Vary the distance between camera and target building between pics, also vertical and horizontal
     5.c If you cannot get ground level pics, take a close up still photo with a drone.

Ensure the still photo does not look similar to any approach videos frames

Approach flights towards target building:
• Length of approach: Rule of thumb here is start with the target building taking up around 5-10% of the total frame and then proceed to fly towards the target building
• Vary your approaches directions, stability levels, camera/zoom levels, exposure, etc... anything to add variability within shots

Capture the location of the target building (map pin or lat/long)
• This allows us to find it on a map for post processing

Also submit srt files along with mp4
• This is useful data for us and most drones should allow for it

We also want IR footage of the same shots if possible
• If pilots have a drone with separate IR camera, or even a combination that does both simultaneously (like DJI mavic 3 thermal advanced) this is highly preferred.

Sample Angle Approach:
https://www.youtube.com/watch?v=_A5_U9-Phh0

Pilot Skillset and Operational Considerations
Ideal pilots should be comfortable and experienced in flying drones, particularly with platforms like the DJI Mavic series. While advanced technology isn’t required, pilots should be able to utilize different zoom levels for varied optical inputs. A key consideration is navigating potentially crowded or complex environments, such as industrial areas with power lines or other structures, and understanding airspace regulations. While the core task involves manual piloting, the ultimate goal is to train a computer vision model that may support autonomous flight in the future.

Location Sourcing and Permissions
A crucial aspect of the project is location sourcing. While the preference is for industrial sites like oil refineries and chemical plants due to their visual complexity and relevance to the technology, these locations often require specific permissions.  Coordination and obtaining necessary permits will be essential, if required.

The scale of the project is significant, aiming for 100 unique scenes per month, and we will have continued need for more pilots.  This volume requires a broad geographic distribution to ensure variety in building types and infrastructure.