Skip to playerSkip to main content
  • 2 days ago
Transcript
00:00How computer vision learns to read the streets.
00:03Graffiti is a planned, creative craft, not random paint on walls.
00:08Vision systems compare before and after images to spot fresh tags quickly.
00:13A fresh tag on a station pillar or the side of a train can appear between two commutes,
00:17and by the next day it might be covered or replaced.
00:20That's where modern vision systems help.
00:22At the core of a memorable piece is letter form.
00:25Caps and nozzles change line width and texture, from hairline cuts to fat fills,
00:30so a single can can produce outlines, fills, and soft blends.
00:34Many pieces are planned, yet the best still leave room for improvisation responding to
00:38drips, wind, or the way a seam breaks a letter's spine.
00:42This Budapest graffiti mural depicts Bender, the iconic robot from the animated series Futurama,
00:48reimagined in a bold street art style.
00:51That tension between planning and improvisation is exactly what makes their mark so visually
00:55distinct, and why they're such strong signals for cameras and algorithms later.
01:00At street level, the winning recipe is change detection plus object detection.
01:05New images are aligned to those baselines using structure from motion and orthophoto generation,
01:10so yesterday and today, line up pixel for pixel.
01:13A widely used approach trained on facade imagery reached 88% accuracy when detecting markings on
01:19orthophotos, helped by a faster R-CNN backbone.
01:23And a carefully labeled set of 1,682 markings across 1,022 images.
01:29Two Wiens Indigo Work designed a change detection data set with 6,902 image pairs collected over
01:3529 sessions along a 50 meter test site.
01:39As one research paper summarized, with an accuracy of 87% and a recall of 77%, the results show
01:46that the proposed change detection workflow can effectively indicate newly added graffiti.
01:52Large transport trials have shown that AI-assisted CCTV can generate tens of thousands of real-time alerts
01:58reports over a year at a single station.
02:04Now, we're not to mention, we're just now to mention a lot of pixels that are meant by selecting the
02:07filters, so we've seen that in that area of setting, if we can Power joyful control.
02:10And, here we are experts at every level of focusing on the machine.
02:12So that's what we are experiencing to be the same, the movement ofaltung, this is a specific
02:15source of our telescopic system.
02:17So, when we are looking at the same purpose, we're looking at what we're looking at.
02:19So, you can look at the same ones, you can look at the same amount of sensors that are
02:24in the same surface of a mechanical load, which can be used for this function,
Be the first to comment
Add your comment

Recommended