I wonder what their FPS requirements are? Most analytics I've seen that work on an "advanced" level usually require a fair amount of frames per second, like 8 or more, and that begets the question of how upload bandwidth might impact the analytics, or if it trades off to low res, low quality highly compressed streams to upload enough frames.
Ex Googler Startup, Camio, Targets Surveillance
A team of former Google engineers are targeting the surveillance market with a startup focusing heavily on analytics and search. In this note, we examine Camiolog's product positioning, comparing it traditional surveillance and rival new entrants like Dropcam.
Luis, here are their rules:
- Maximum 1 frame per second for jpeg snapshots sent from any camera.
- Maximum 1MB per jpeg snapshot.
- Maximum 2MB per individual video clip (typically, 10 seconds of video+audio at 15 frames/second in 640x480 resolution).
- All cameras must upload only upon motion detection.
- Your account covers all cameras in a single residential location (i.e. all cameras on the same home router).
Basically, the service depends on the camera first detecting motion events (max 15fps / SD) and then sending those clips to the cloud for (further) processing.
I should have figured that- camera side detecttion. I kept thinking of it as a wholly cloud solution with cloud based analytics.
Could this be used in forensics and court of law? From what I understood it is literally making snapshots out of the video and then using that to do the analysis. It sounds a lot like the technology in Picasa which analyses your stored photos and brings out all the faces. Once you put a name to the image it then takes care of the rest.
That's a good question / point. I am sure it can be 'used' but there is a higher risk that it is challenged as the 'whole' video is not available. One could then argue that the system missed critical parts that would validate a defendant's innocent. Right now, it's just for home use, so it's probably not a big deal but the whole approach does increase risk of acceptance in court.
From a residential viewpoint, the ability to find out which neighborhood dog made a deposit on your lawn or who approached your front door while you were away are great reasons to have the ability to easily search video.
There are many activities that occur when you’re not watching live or at an unknown time. The ability to search in a specific area of a scene, search by size or shape, search by audio level, etc. greatly improves the time and hassle it takes to find what you want on video. Recording and searching simply based on motion detection doesn’t allow you to quickly find what you’re looking for.
I would think that ease of search functionality, whether the technology is at the edge, in the recorder or in the cloud is important in professional surveillance applications as well as residential.
If search is so important why is Avigilon worth 50 to 100x more than 3VR? :)
Now, that by itself, is not conclusive. But ask yourself - what surveillance company that focused on search is now a major industry success? Literally, there is none. That should tell us something.
To your point, "Recording and searching simply based on motion detection doesn’t allow you to quickly find what you’re looking for."
The reality is that (1) most people are not looking for much and (2) most of the time they can get it. It might to take a few minutes instead of a few seconds to see who was at your door when you were away but it's generally feasible.
I saw the problems of selling searchable surveillance first hand. It plays well in the Valley because Google is gargantuan, but it's just not that important for surveillance users. It does not drive many buying decisions and the advantages that search focused providers have over ones that don't are typically not enough to win deals.
Overall, my experience is that powerful search is a 'nice to have' but not a core business driver.
Thank you to everyone who pointed me to this article today. I'm happy to have discovered ipvm.com.
It's a fair critique that Search is not a major driver for consumers today. But we hope to lift expections of what you get from video by enabling cameras to tell you what you need to know whenever/however you want to know it. To consumers, smart video doesn't *feel* like a Search problem (and perhaps we'll always use term "Smart Video Monitoring" instead of "Search" for that very reason), but all the work in feature extraction, machine learning, indexing, ranking, SxS evalutations, feedback loops, etc... is at its core a kind of Search problem. And precise Alerts (which all consumers understand) are just the flip side of Search (since an Alert is just a persistent Search). So the warning about our consumer market/product fit has my full attention and concern. But the daily usage I see with smart search and alerts has me very optimistic that it's an awareness problem (i.e. a Marketing problem to solve)
One clarification to this article is that our setup of a Camiolog-Ready IP cameras is actually easier and faster than Dropcam's setup. Watch the video of an outdoor WiFi camera setup out-of-the-box in less than 1 minute at camioblog: 53-second Setup. No Cables!. There is no need to open network ports or know anything about FTP/HTTP once we know the CGI commands required for a particular camera model. The manual setup for Axis was required only because v5.4 of its firmware omitted CGI interfaces to set motion detection and upload tasks. As long as the camera's firmware has an HTTP GET/POST API interface to make settings changes programatically, it can become Camiolog-Ready easily. Easy out-of-box setup is critical and possible even while remaining open to all camera sources.
Carter, thanks for responding.
On search and alerting, if you can deliver accurate alerting, you should focus hard on that. Alerting delivers far more value than searching for a monitoring / surveillance application for the simple reasons that alerts can stop something bad from happening while searches just tell you who or what did it. On the other hand, alerting accurately is hard and there's a long trail of VC backed companies who have failed at it.
On setup ease, there are a number of steps that Camiolog requires that makes it more complex than the Dropcam approach.
First, what I would consider 'pre-requisites':
- Dropcam's home page call to action is much clearer - buy a camera with a big buy button. With yours, one needs to click on a small 'join now' reference or a still relatively small 'get started' button below the fold. Then you are sent to the 'Get Started' page which has 3 steps, buried in the middle is the critical step of buying a camera (creating an account to record own's cameras is almost certainly a negative experience for all but the most technical and determined users).
- With Dropcam, it's really clear that you are buying a Dropcam camera. With Camiolog, it's confusing to know what a 'Camiolog Ready IP Camera' is. It's a Sharx camera? What Sharx cameras? What firmware? etc.
I think you are losing 80% of people just because it's so confusing / complex to get started online. Sell your own cameras, but a big buy button on the center of the home page.
Then, there's the setup specific things. According to your instructions, one needs to either go to a specific URL or download / install an app. By contrast, with Dropcam all you do is plug the USB cable into your computer and a wizard 'magically' appears. This is a lot easier for non-technical people. They now how to plug devices into a USB port, they tend to get lost reading instructions, finding websites, apps, etc.
Finally, there's the configuration of the system and the layout of the UI. It's hard. Setting up zones. Experimenting with the smart image filter. Figuring out where live view is or how to get to it. When you try things out, the display seems to radically change and it's not clear where one is at.
Dropcam is easy - easy to buy, easy to setup, easy to watch video live and see recent recorded video. It's certainly not as powerful as Camiolog, but for the consumer market, I doubt the end to end ease will trump power for most every user.
I got no love for Dropcam, and I would prefer a stronger, more dedicated engineering organization to 'win'. But Camiolog needs to work hard on making it easier for people to get started (from buying to install to basic use).
Completely agree with your points John. We're doing a major UX refresh that we had postponed until we had completed work on our API. Now we're underway with site and app refreshes that address the exact points you're making.
I mistakenly put UX on the back-burner because I so badly wanted to prove to myself that the image processing pipeline would work at scale for an accessible price across all camera sources. Even my most technical advisors were concerned that our real-time cloud processing might not be feasible. Today's sad Mandela news reminded me of his “It always seems impossible until it’s done.” In any event, we're excited to launch the new on-boarding and app UX soon.
This is awesome. Great article.
I believe that Search will be the next big thing in Video Survillance industry. John, you have an article with interviews of security personnel who told you that most of the time they spend to investigate the case. One case can take 4 hours (as I remember from article), just to find that moment from archive. So, real challenge is not about preventing the case because of super real time analytics. It is about saving money on spending less resources on investigating. I also read some article about London police when they reported that most of the cases they are not investigating using video footage because it is too expensive. So, I believe, that inability to search - is N1 problem of our industry. And that one who will offer reliable, accurate and easy to use solution will get the market. You can see that our market leaders are growing slower than market itself. It clearly shows the demand for more advanced technologies.
Your evidence is an IPVM case study?
Listen, there have been people who have complained about search performance for years. This is not new. And despite those issues, no one who has led with search has been successful.
And how do you propose to solve the London search problems? Tens of thousands of cameras from hundreds of different systems using myriad types of cameras and recorders? There's no magical search solution that solves it.
You have two problems here - (1) search is NOT the top priority, alerting / stopping events is - always has been and always will be (i.e., do you want to get an alert on a terrorist about to drop a bomb or find out who bombed afterwards? The former is always far more valuable) and (2) the enhancements to search have not been so dramatically better to radically improve performance.
Getting an alert on a terrorist about to drop a bomb is magical, but finding out who bombed afterwards is a real job which can be improved. Why this technology is not market driver, because it is too expensive: you need to decompress all video streams and analyze on server. But now there is a way to do scene analysis on camera side and store/search on server side. Let's see what will happen during next 3-5 years. I doubt if anyone will deploy system without search capabilities.
Murat, the reality is almost all security users prioritize their spending on alerts. And, by the way, clicking search and finding a terrorist in the city of London is just as magical as alerting against a terrorist about to drop a bomb. The difference is the alert can prevent disaster, but the search can only more quickly document the damage.
Who's doing 'scene analysis on camera side' with what cameras, at what costs, with what capabilities? Because from what I have seen with color, people, face, motion analysis, searching will still be very rough and not that radically better than the search capabilities that are already built in to VMSes.
John, is any of this VMSes has built-in real time analytics to detect some threats ? But all of them has different approaches to improve search. So, this is definitely a trend. I don't understand why you don't accept this fact. Do you see any other way to improve VMS ? Viewing, recording, exporting, integration...I see only search/investigation and may be viewing as an old analogue style of showing all cameras in MxN mode is boring :)
Murat, you claim that "Search will be the next big thing in Video Survillance industry" and "that inability to search - is N1 problem of our industry". I certainly do not accept either of those propositions.
If you simply want to say that companies can and will incrementally improve search, sure, no objection, but it is not going to be the 'next big thing' or solve the industry's biggest problem.
I just don't see any bigger problem. And I believe that there's search solution that solves: "Tens of thousands of cameras from hundreds of different systems using myriad types of cameras and recorders"...in principal, of course. But very soon it will be typical, like thumbnail search.
Murat, well then you should look harder...
There are solutions 'in principal' to many problems but making them do so 'in practice' is the far harder thing. If you want to convince me you are going to have to speak in far more detailed specifics about your solution.
John, I just realized that many things happened this year from camera side and I'm sure that it will change our market from technologies perspectives. Few years ago, only Bosch offered camera scene description metadata from camera. Now Sony and Panasonic. Beginning of next year we expect it from Samsung and later from HikVision. But what current VMS can do with this data? Nothing. This is interesting to predict what features will be available with this new approach and which of them become MUST for any VMS. I'll be happy to share my vision in separate thread. BTW, what about allowing manufacturers to have their own thread, kind of blog to share technological ideas and to discuss them with folks ?