welcome to FI3LD
Sports events digital content automation for everyone
Bib numbers are a tool for race directors to tell these thousands of runners apart and to simplify the seemingly impossible task of getting a time for each and every one of these runners. Bib numbers have many event applications, including:
→ Road Races
→ Track and Field Competitions
→ Cross Country Races
→ Biking Events
→ Baseball, Hockey, Soccer or Football Tryouts
→ Dance Competitions
→ Any other event that you need to differentiate people by way of a highly visible number.
Small facts about marathons
3% to 5% of the worldwide population runs and 1% have run a marathon, 0.22 have finished one.
>1000 marathons are organized every year with a record breaking 51388 participants, as the largest participants for a run was for “A run for the Pasig river” in Manila Philippines
Obstacle course and other themed races
Obstacle races or (OCR) have differentiated themselves from classic runs and attracted an entire new audience.
Our mission first mission is to cover a marathon and output a single personalized video for every runner within the same day of the event.
Those videos will feature a montage of each runner recorded from different cameras installed along the run, the videos will present cinematic technique such as pan in/out, zoom in out all centered on the runners. And this for every runner. The length target for each video is at least 10 secondes with at least 4 scenes.
All starts with high quality 4k resolution video shot with a wide angle setup and fixed position, this allow us zoom in the footage and render sub frame in full hd. As well it gives additional data for the image classifier and text extractor.
We run our first AI layer to detect BIB in the video
This is a multi step process, we first estimate the bib readability, look for the bib corners, correct the perspective and run a text extractor.
Tracking and log
We log the coordinate of every bib on every video frame using a combination of BIB tracking, face tracking and body tracking.
The tracking data is used to create sub region in the video feed, each sub region is attached to a confirmed BIB detection
Each sub region is rendered and attached to its BIB, different effect can be applied such as pan-in, pan-out, zoom in…