Video recognition requires intensive processing by your computer. The minimum requirements for a single camera at QVGA are:
- Windows XP Service Pack 3 and newer, including Vista, 7, 8, and Server releases
- 2.0GHz Pentium 4 processor or higher
- 2GB RAM or higher
- Mac OS X 10.8 or higher
- 2.0GHz Intel-based processor or higher
- 2GB RAM or higher
- Minimum screen resolution of 1024 x 768
- At least 50GB of free disk space per camera is recommended
- Allocate more disk space to record higher resolution or more temporary video. Click here for details.
- Video can be stored on an external drive
- Higher resolutions require more disk space to store equivalent amounts of video
- An internet connection for software download, updates, and activation.
A USB webcam or IP camera is required. IP cameras require a network connection to the computer where the software is installed. Sighthound Video is pre-configured to work with many webcams and network (IP) cameras. Click here for a list of supported cameras. Note that recognition performance may vary by camera. Axis cameras, for example, provide some of the best results.
How many cameras can I run?
Video encoding and decoding and computer vision analytics requires intensive processing by your computer. A fast quad-core computer can potentially run up to sixteen cameras, but an older computer may struggle running one HD camera.
The main factors determining how many cameras can run on a given computer are:
Speed and number of cores of your computer. Processor speed alone doesn't necessarily specify how fast your computer runs programs. Equally or even more important is the number of "cores" on your computer. A multi-core processor contains more than one independent processor on the same chip. A dual-core processor is like having two computers in one box (and a quad-core is like four). Netbooks and the most inexpensive computers typically have the slowest processors (e.g., Intel Atom), and may have difficulty running one camera. A site like http://cpubenchmark.net/ can give you an accurate speed comparison between processors.
Resolution at which video is received. A low resolution for streaming video is QVGA (320 x 240). Higher resolutions, such as 1280 x 1024 requires processing seventeen times the number of pixels. This places a much heavier load on your computer.
H264 Streams. Incoming video from cameras that are not H264 must be decoded and will significantly increase the resource usage.
Activity in the scene being recorded. The more motion in a video, the more processing is required to analyze what is going on. A camera pointed outside at a busy street will required considerably more processing than an indoor camera in a room with no people, or an empty backyard or parking lot. Note that very dark scenes could generate video image artifacts and noise that make them equivalent in complexity to "busy" scenes.
Other applications. The more you need to use your processor for other applications, including automated programs like virus scanners, the fewer cameras you can run on Sighthound Video.
In general, you should be able to run four to eight cameras per core on a dedicated machine (a little less if you are also using the machine for other applications). For example, a fast dedicated quad-core machine would be expected to perform as follows:
- VGA video, low to moderate activity: 32+ cameras
- 1280 x 720, low to moderate activity: 22-30 cameras
1) A 2013 mac mini with a quad core Intel Core i7-3615QM 2.3GHz CPU and 16GB memory uses about 40% of the CPU when monitoring a mix of 9 cameras with resolutions of 1280x720, and moderate activity
2) A 2015 model of Dell XPS: Core i7-4790 3.6GHz CPU, 16GB of RAM uses about 24-30% of the CPU when monitoring 10 cameras (9 at 1280x720 and 1 at 1920x1080), and low activity
3) A Windows 8 laptop with an Intel Core i3-3227U 1.90GHz CPU and 4GB of ram uses approximately 34% of the CPU monitoring 6 VGA (640x480) cameras, low to moderate activity
4) 2010 Gateway DX4320 with AMD Phenom II 945 (3GHz) CPU and 6GB RAM uses around 33% CPU when monitoring two 1080p cameras, two 720p cameras, and one 480p camera with low activity
Note that these results may vary accross different video input sources (codec, frame quality, etc...) and scene activity levels. These examples all used H264 cameras - using older cameras that do not support H264 streaming will increase the resource usage.
For more tips on optimizing performance, see additional performance tips here.