Een drijverstuw benedenstrooms van een weidevogelgebied in Pijnacker. Met de schotbalkstuw aan de zijkant kan handmatig het peil worden gereguleerd. Bij het hoogste peil (broedpeil) en bij zware neerslag kan ook stroming over de drijverstuw plaatsvinden. Als het benedenstroomse peil stijgt, stijgt ook de drempelhoogte van de drijverstuw, waardoor tijdelijk water in het vogelgebied kan worden vastgehouden om de belasting van het benedenstroomse gebied te beperken.
* English *
A float-controlled weir located downstream of a meadow bird habitat area in Pijnacker. The stop-log weir at the side allows the water level to be regulated manually. At the highest water level (breeding season level) and during periods of heavy rainfall, water may also flow over the float-controlled weir. As the downstream water level rises, the crest level of the float-controlled weir rises as well, allowing water to be temporarily retained within the bird habitat area, thereby reducing the hydraulic load on the downstream area.
* Technical data *
Processed exclusively with Spirula Studio on a 5070 desktop with the following settings:
- Recording -
Recorded on a cloudy day with a light drizzle and low wind speed with an Insta360 X5 on a 1.2 m selfie stick.
2 minute 8K video with 400 ISO, auto shutter speed and 5000 K white balance.
Three passes at each side of the weir at varying heights and a loop above the water line and overhead across the path. 1 Lens faced down when possible to minimize raindrops on the lens.
- Preprocessing and SfM -
Fisheye frames extracted at 2.3 fps (622 images), settings "maximum quality"
SAM 3 masking for person, backpack, license plate
5% fisheye edge static mask
80% of images were aligned, missing images were frames looking at the sky.
Preprocessing and SFM took 15 minutes in total.
- Training the splat -
Default 360 camera - High settings (3M splats, 50k iteratations, SH 3)
Bilateral grid and PPISP left ON to compensate for automatic shutter speed and whatever else PPISP helps with
Mask edge extension changed from -0.005 to -0.002 to prevent masked areas from being too big.
Training took 2 hours. Progress was still made until near the end, especially in the distance, so the number of iterations seemed correct.