2017 Video Surveillance Predictions
Read the full report here
Let's compare member predictions from 2012, which were for 2017, which is now here.
The chart:
Video analytics is still not a big thing but let's see if 2017 will be the breakthrough, still skeptical since it takes time for new technology to get adopted and we do not see any breakthrough mainstream video analytics this year.
And I got the next big thing very wrong in 2012, predicting:
By contrast, the runner up, edge storage, has a much clearer shot at impacting the market.
Back then, HD-SDI was doing terribly and edge storage looked to be the way to get lower cost small systems. But then HD analog started to appear the next year and changed market dynamics.
Also, I'll say there is a good chance that we see Hikvision and Dahua start to pull back / slow down spending internationally as (1) they realize their approach is economically unsustainable and (2) slowing China domestic market makes them more conservative. Tactically, though the issues are different for the two; Hikvision has reached a point in NA where growth is getting much harder because of their large size and opposition at the high end; Dahua NA has a very low base to start but is incredibly poorly organized. They should try to hire the Hikvision NA president Jeffrey Hu.
In terms of video analytics, a breakthrough doesn't seem likely, even on the horizon.
Considering that even high-powered CPU clusters still show to have limited analytics success, and there are fundamental physical obstacles at play with camera mounting and FoV limitations, it's not like the market is waiting for super high-end, high cost tech to become cheaper or small enough run on camera chipsets.
It seems the 'breakthrough' has yet to even be seen at the development stage.
It seems the 'breakthrough' has yet to even be seen at the development stage.
The deep learning proponents disagree. The question becomes is how soon to a deep learning mainstream video surveillance camera / recorder and how much better will they perform?
I agree. To me Deep Learning still holds great potential. I am cautiously hopeful. The big concern is that the Western manufacturers do not seem to be on board... at least publicly.
If they could simplify the searches into more common google style searches for events this could save significant time for guards.
It would be nice to see analytics do something more impressive with live video, but the processing power required would undoubtedly be mammoth.
Analytics faces just as many challenges on the human factors side of things, i.e., the need in critical analytics applications (like knowing when someone is dangerously close to a train track or about to fall into a subway track) for a human being to actually view the alert to verify the action required. Is it a real emergency, kind of an emergency, almost an emergency, etc.????
Now, the problem I am seeing is security/operations staff needing to review/respond to maybe 5-6 alerts an hour per analytic point and seeing this as too obtrusive on their existing work tasking. Being that most of the alerts are not going to be real emergencies, they become jaded to the task and begin to complain that there are too many false alarms. Expectations that every alert is going to mean something is not realistic in many real-life scenarios where there is heavy human and/or machine activity.
Who/when will it be a guaranteed that when using self-learning, server-based analytics that only real alarms will be reported in the system?
I suspect that the deep learning element will not be limited to learning their own environment, but others as well.
Multiple cameras will provide multiple perspectives of an object with the idea of rendering a 3D understanding of the object/environment using data from each camera.
Processing power will continue to increase in speed and decrease in size. I think this will help empower the 'learning' aspect even further. Probably first server based, then edge next.
Multiple cameras will provide multiple perspectives of an object with the idea of rendering a 3D understanding of the object/environment using data from each camera.
Matt, have you seen any commercial vendor offer this yet?
This is only speculation (with a bit of wishing thrown in). I haven't seen any offering like this.
Axis has been working on 3D for several years, I can't say if it is still in the works???
We expect to get incredible projects just because of archive search analytics. This is definitely the next big thing. Face search, color search, tracking search, etc...Not many VMSes support it and CPU resources needed, that is why it is not main stream today. But tomorrow for sure it will be famous.
"China will push back. They'll apply their thousands of engineers to settle the issue, and they'll produce products that are objectively safe from a cyber perspective."
Someone hasn't figured out yet that China is intentionally making insecure products to use as tools in their cyber-war against the West. Bless his heart....
"They'll apply their thousands of engineers to settle the issue"
On the positive side, Hikvision and Dahua's fallacious marketing of engineer count = technical competence is working at least on some integrators.
Cloud:
Cloud solutions will make integrators and distributors less and less relevant as the solutions get simpler to deploy and manage.
Ask questions and get answers to your physical security questions from IPVM team members and fellow subscribers.