KIO OutdoorExtended Test — 14.12.2025

KIO OutdoorExtended Test — 14.12.2025 Hardware: • Camera: GoPro Hero 12 Black action Cam x3 • Selfie stick: 2.5 M • Apriltags Software: • Training Pipeline: Reality scan for Colmap data, Jawset Postshot 1.0 , Splat 3 Workflow: The capture path was walked once in a zig-zag pattern using three cameras mounted at three different heights. This was followed by an additional walk around the sorting container and the transporter. AprilTags were used for scaling and orientation. Camera capture was triggered simultaneously on all three cameras using a Bluetooth remote app. • Height: ~2.6 m • Height: ~1.5 m • Height: ~0.6 m Dataset: • Total number of images: 1100 • Date of Capture: 14.12.2025 • Time to Capture: 6 Minutes • Operator: A.B. Key Observations: • Using three cameras at different heights simplifies data capture; however, careful path planning is essential. Missing data was observed on the right-side path. Adding an extra capture path parallel to elevation changes helps avoid gaps while also reducing excessive overlap mistakes , especially in visually similar areas (e.g., curbs captured by the lowest camera). • Voice commands were abandoned due to reliability issues. A third-party Bluetooth application was used instead to start and stop capture on all three cameras simultaneously. • Using RealityScan to register camera positions and COLMAP data significantly speeds up the training process. • Both RealityScan and Postshot can be used to detect AprilTags and to correctly scale and orient the resulting Gaussian Splat output. KreativInstitut.OWL @kreativinstitutowl