Can your system recognize a person being carried? If so, then can an alert be generated?
In light of the recent rape case at Vanderbilt, this would be a great feature. Most universities have hundreds, if not thousands, of security cameras. The security force can't monitor all of them. Motion detection helps but isn't enough on a busy campus. Having something that detects suspected foul play would help. Carrying a person would be one of these flags. A person with a gun (happened in a UAB parking garage) could be one.
Detect suspected foul play
Moderator: Staff
2 posts
• Page 1 of 1
-
tesla8er - Posts: 30
- Joined: Sat Jun 28, 2014 3:17 pm
Re: Detect suspected foul play
Hello rturrentine,
No, I'm afraid Sighthound Video can't distinguish someone being carried from a person walking normally. That level of detail (not is it a person, but what are they doing) is the subject of various research projects into action recognition. I think it is fair to say that the state of research isn't at the level where it would be considered accurate enough for commercial use.
On the question of a person with a gun, there are products being marketed to agencies and institutions which claim to do that. Product reviews we've seen suggest that they don't work too well. It's not surprising - both of these are hard computer vision problems. Even success rates of 90% might not be that useful.(Would a 10% false alert rate of "Man with Gun on Campus be acceptable? Probably not.)
As camera resolutions get higher, the opportunities for more sophisticated computer vision increase, but Sighthound Video is not there yet.
Regards,
Stephen
Sighthound
No, I'm afraid Sighthound Video can't distinguish someone being carried from a person walking normally. That level of detail (not is it a person, but what are they doing) is the subject of various research projects into action recognition. I think it is fair to say that the state of research isn't at the level where it would be considered accurate enough for commercial use.
On the question of a person with a gun, there are products being marketed to agencies and institutions which claim to do that. Product reviews we've seen suggest that they don't work too well. It's not surprising - both of these are hard computer vision problems. Even success rates of 90% might not be that useful.(Would a 10% false alert rate of "Man with Gun on Campus be acceptable? Probably not.)
As camera resolutions get higher, the opportunities for more sophisticated computer vision increase, but Sighthound Video is not there yet.
Regards,
Stephen
Sighthound