Testing Vitamin D Video Analytics SoftwareBy Scott Oie, Published on Mar 05, 2010
With significant tech blog coverage, Vitamin D has generated widespread interest in using their video analytics. Two of the most commonly cited points is their technology, called HTM, that is based on a "new theory of human intelligence" and their "freemium" business model that allows for free or inexpensive on-line access to their video analytics.
In this test, we examined Vitamin D Version 1.0, running on a Windows 7 laptop analyzing video from 3 cameras: a Logitech webcam, an IQinVision basic series camera and an Axis HD camera. We choose a range of cameras to better understand how Vitamin D performs on different inputs. The Vitamin D software Starter Edition can be freely downloaded so you can test for yourself.
Our key findings include:
- Easy to set up and worked with a wide range of cameras
- Significant false positives and negatives in certain environmental conditions
- Provides video management (record and search) capabilities (though quite basic)
- Performance can vary greatly based on capabilities and characteristics of camera used
Vitamin D is software designed to run on Windows or MAC PCs and connect to webcams or regular IP cameras (see the specification sheet for background). Vitamin D works with a modest range of network cameras including Axis, ACTi, Panasonic, IQinVision (see camera interoperability specification).
Vitamin D has two major functions: (1) analyzing video and (2) providing video management of that video.
For analyzing video, the system scans for objects and people on cameras by default and without configuration/optimization (though advanced users can change the minimum pixel size or set more complex rules to fine tune detection and alerts). Analyzing video happens as a background process (not a service) and runs regardless of the Vitamin D client.
The Vitamin D client allows users to (1) watch live video, (2) search for specific events and (3) configure the system. Most importantly, searching for specific events is done via a collection of events. Each event is listed in a row with the type of event (e.g., person) and a small thumbnail of the image for that event). Users can scan through the rows and play back video from each event.
The Vitamin D client is a local client only and does not allow for easy web based or thick client viewing from a remote computer.
The system also allows events to be emailed or dispatched to users. As we examine in the performance/usability section below, common conditions can cause extreme numbers of false alerts. We recommend users set up more advanced rules and only send notices based on those rules (to eliminate messages triggered by common false alerts)
By default, the system records 48 hours of continuous video. After 48 hours, video is discarded unless the video is part of a specific event captured by the system. This cannot be changed.
The maximum recording duration is set by the amount of hard drive space one allocates for storage. By default, this is 25 GB but it can be changed in the Vitamin D client.
Vitamin D offers 3 product packages that differ in maximum resolution and number of channels supported.
Vitamin D can be acquired and purchased on-line:
- The entry level offering is free but only supports 1 camera and 320 x 240 resolution.
- The mid level offering is $49, supports up to 2 cameras and 640 x 480 resolution.
- The high end offering is $199, supports unlimited number of cameras and 640 x 480 resolution. Unlimited support is the licensing agreement for a given server. However, Vitamin D is fairly processor intensive and is unlikely to run more than 4 -8 cameras even on a new 2009/2010 machine.
Our tests concurred with Vitamin D's system hardware specification recommending 2 cameras per CPU core on a dedicated machine (1.5 cameras per CPU core on a machine multi-tasking with personal applications): In our tests, we used a PC with an Intel Core 2 Duo CPU and connected two IP cameras and one consumer webcam.
With three cameras connected simultaneously at 320x240 resolution on a dedicated machine and with the Vitamin D client not running, the CPU usage was measured at the range of 55% - 60%.
Changing the resolution of the three cameras to 640x480 resolution, CPU usage risen to the range of 75% - 80%.
In all tests, we did not run any other non-essential applications (web browsers exited, video players shut down, etc.). With the client application running, CPU utilization increased an additional 5-10%.
Setup and Configuration
In the screencast below, we examine key points in setup and configuration including:
- Associating cameras/webcams to system via a wizard is beneficial for non technical users.
- Integrated IP finder function included in camera setup wizard but not accessible in normal interface.
- Rule editor interface options logically organized but setting up detection regions may be confusing for non technical users.
System Usability Challenges
After placing Vitamin D's analytics under extensive environmental test conditions, we found the following conditions provide the best performance for the system.
Recommended best conditions to apply analytics:
- Indoor scene
- Still condition (no vegetation or curtains that may move from a gust of wind)
- Small to moderate field of view
- Controlled daytime lighting
- Analytics affected depending on the output camera noise of different cameras.
- Experienced poor low light performance unless subject/object produced its own light source (holding a flashlight, car headlight turned on). [You could require lights to be on at night but this may not be practical or wanted, etc.]
- High amount of false alerts from shaking trees, moving curtains, stray lights, and light camera shake.
- Subject not detected if scene has a field of view width beyond 40 feet.
Contrast to 'Smart Cameras'
In the professional surveillance market, 'smart' cameras or analytics running on-board cameras is the 'hot way to go' (though more analytics are still deployed server based). Since VitaminD is a server based solution, it's worth considering the trade-offs.
On the plus side, VitaminD can work with many cameras, including cameras that already deployed or low cost cameras that can be bought from electronic/computer retailers. On the negative side, VitaminD's performance can be heavily impacted by the camera chosen. Users may not realize this and will blame the system regardless of who's fault this is. We saw this practically with the complete darkness of the webcam at night (missing almost everything) and the noise from the IQ camera (constantly false alerting).
VitaminD can reuse existing PCs one already has. However, because VitaminD is so computing intensive, 4-6 cameras can max out a $500 computer. While the software may be very inexpensive, the hardware costs can become non-trivial for more than a few cameras.
Because various combinations of cameras and VitaminD can cause different problems, troubleshooting and support can be more difficult. [VitaminD does not disclose its phone number on the website] You may get email or forum based support but it may be hard for them to determine specific causes from different cameras.
On the other hand, smart cameras at a consumer or SMB price point are rare or non-existent. While ioimage, ObjectVideo (through OEMs), VideoIQ, etc. offer smart cameras, they all are $1,000 or more. The only 'lower-cost' smart camera we recall is the recently released Archerfish Solo which is about $400 but requires a monthly subscription fee.
Recommendations on Use for Residential or SMB Applications
We think use for residential and SMB is promising but there are important constraints for these users to keep in mind:
- While the analytics are easy to setup, it's easy to place them in environments where they may not perform well or with cameras that may not be sufficient for the lighting conditions.
- While Moore's Law makes computing power cheaper every year, right now, VitaminD takes a heavy load on commonly available PCs. If you need more than 4 - 6 cameras, this system is probably not practical.
- The VitaminD VMS, outside of intelligent searching is very rudimentary. If you need remote live viewing, PTZ controls, extended continuous recording or other common VMS functionalities, you are unlikely to be happy with VitaminD (at least version 1.0).