Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Face / License plate recognition is erratic #162

Open
alex9099 opened this issue Jun 1, 2018 · 16 comments
Open

Face / License plate recognition is erratic #162

alex9099 opened this issue Jun 1, 2018 · 16 comments
Labels

Comments

@alex9099
Copy link

alex9099 commented Jun 1, 2018

Hi, i was viewing my processed track but the Face / License plate recognition is awful, lot's of thing that are not faces or license plates are blurred, can't the images be reviewed by it's owner? Also can i download the whole track without the blur(assuming that i'm the owner of the photos)?

@mds08011
Copy link

I agree. It would be beneficial to have a method for identifying images with incorrect blurring.

For example, I have a photo where a bike route sign was blurred during the image processing. (Sign on right side between the gap in parked cars): http://openstreetcam.org/details/1263209/55/track-info

@alexandrui-telenav
Copy link
Contributor

We're working on a machine learning approach ( the one on the market is an open source algorithmic solution ) that will improve this and add support for user reviews.
It's going to continue to be awful for a while unfortunately..

@alex9099
Copy link
Author

Also manual tagging would be nice :)

@mds08011
Copy link

Thanks for the update @alexandrui-telenav

@jsniff
Copy link

jsniff commented Sep 4, 2018

@alexandrui-telenav @mds08011
Our company helps other companies with their privacy protection needs. We have rly good face and license blurring software. Would either of you guys be interested in seeing a demo? We built it for the body cam industry originally and now looking for new applications. Would be interesting to see if we can help you guys.

@mds08011
Copy link

@alexandrui-telenav In the meantime, I think it would be beneficial to allow the sign detection to override blurring. I have run into some instances where the Sign Detection accurately identifies a sign, but is then blurred over, making it unhelpful for mapping purposes. This seems to mostly happen with highway signs.

Example 1: http://openstreetcam.org/details/1289493/20/detections
Example 2: http://openstreetcam.org/details/1289493/44/detections
Example 3: http://openstreetcam.org/details/1287947/3/detections

@alexandrui-telenav
Copy link
Contributor

Fully agree @mds08011 . The Machine Learning component we developed shows promising test results ( not yet integrated in production ) and potentially allow what you're suggesting. Current architecture is not flexible enough for that.

@mds08011
Copy link

That makes sense. Thanks for the update!

@jsniff
Copy link

jsniff commented Sep 18, 2018

@alexandrui-telenav @mds08011

Our company helps other companies with their privacy protection needs. We have rly good face and license blurring software. We've done a lot with body cam companies: http://www.suspecttech.com/ Would either of you guys be interested in seeing a demo next week? What works?

@mds08011
Copy link

@jsniff I am not associated with the development team. Given that a solution is already being worked on internally, I am guessing they are not interested in outsourcing this work.

@jsniff
Copy link

jsniff commented Sep 18, 2018

@mds08011
Malcolm understood. The thing is, having seen so many projects done by large companies, they almost never seem to end well, unless it's their core expertise. Ultimately we're hoping our clients save money by spending less than they would to build and also by delivering a very high quality solution that actually works well in production. As you can see by the current setup, it's not very good to say the least.

@alexandrui-telenav
Copy link
Contributor

@jsniff thanks for the offer, you have a good point but we already trained a ML model that shows very good results and integrates well with the current processing pipeline.

@jsniff
Copy link

jsniff commented Nov 4, 2018

@alexandrui-telenav Alex, understood. we have a rly, rly robust platform on cloud and on-prem. Most people think they have a solution and their stuff isn't very good. Here’s some initial materials below; can we show you a demo next couple weeks?

Short demo videos:
https://www.youtube.com/watch?v=o03MtrS-fsg
https://www.dropbox.com/s/op6pspzgzm9inx3/hispanicspeedup.mov?dl=0

@jsniff
Copy link

jsniff commented May 27, 2019

@alexandrui-telenav Alex, are you guys happy with your new technology? Let me know if you'd like to revisit discussions.

@alex9099
Copy link
Author

alex9099 commented Jun 7, 2019

It seems to still blur things that it shouldn't for example here, at the distance everything is blured https://openstreetcam.org/details/1469714/0/track-info

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Projects
None yet
Development

No branches or pull requests

6 participants