ORB-SLAM 360° Point-Cloud Mapping

We need a way to accurately document the interior of construction sites — specifically the walk path taken through the building and a 3D point cloud of the surrounding space — using footage captured from a helmet-mounted Insta 360° camera. The process should be fully automated, so that after each site visit, the captured footage can be processed into the required outputs without manual intervention.

The camera's position and path through the building, as well as a 3D map of the space, must be derived entirely from the camera's visual footage and its onboard motion sensor (IMU) data, with no dependency on GPS.

Available Inputs
• Insta360 X4 camera, mounted on a construction helmet
• Raw INSV video files recorded during each site walkthrough (Insta360's proprietary container format)
• IMU data (accelerometer + gyroscope, ~200 Hz) embedded within the INSV file

Expected Output: The primary output is a walk path pose file — a precise, frame-by-frame record of the camera's trajectory through the site, capturing its position and orientation over time (timestamp, x, y, z, and orientation as a quaternion). Pose file should be perfectly scaled. The secondary output is a 3D point cloud of the interior environment, provided at two levels of detail: a lightweight sparse cloud for quick preview and verification, and a dense, metrically accurate cloud (millions of points, real-world scale). Both outputs for a given walkthrough should be aligned to the same coordinate frame.