Home
BYOC is an RK3066/RK3188 based system that turns an Android Google TV player into an intelligent camera system. You can of course just use it as a wireless IP based webcam as well. I've learned a lot since implementing my first cameras over a year ago. Now I'm building my own kernels and Ubuntu distributions. I have improved my OpenCV installation script to build almost everything from source. I have added Java to the mix with Python, so you have a choice of bindings. You can still just build your own camera without any programming skills, but I give those so inclined the ability to get a Computer Vision system up on dual or quad core mini computers. I will continue to improve scripts that monitor network connectivity and detect USB camera failures.
Some of the advantages of using a self contained intelligent camera system are:
- An attached USB camera is more reliable than most wifi cameras as it relates to connectivity and FPS consistency.
- A wider selection of camera resolutions and features for the money. You can build a high definition quad core camera (1280x720) for less than $100 US (It's around $75.00 with free shipping). You can also choose the USB camera that fits the application at hand. Swap out the camera for a fraction of an IP based equivalent.
- Resilience in terms of network connectivity which isn't required to capture video. Once network services are restored the video files can be transferred via email, SCP, etc. Most IP based cameras do not have storage hence you lose the video stream during the outage. Your BYOC camera can last hours or days without network connectivity!
- Distributed processing saves you from having a system powerful enough to capture all of your camera's video streams and process them. The camera can run the same software as a full size system, so in essence you do not need a central PC for processing video streams.
- Being based on a popular Linux distribution allows for easy customization instead of the crappy Chinese software on most cheap IP based cameras. You can even load a GUI desktop for management if desired.
Here's an example of optimized network performance compared to always streaming cameras. The CVP server uses a centralized model with four 640x480 MJPEG cameras that operate an average of about 15 FPS. During night (least amount of traffic) time hours incoming traffic is just above 4 Mbps or about 1 Mbps per camera. This is because a centralized server is reading frames all the time. The spikes you see are SCP activity from the BYOC cameras.
Now look at Front camera which is 1280x720 (3 times the resolution of 640 x 480) at 10 FPS. Note the outgoing traffic averages 252 Kbps. So that means about 1/4 the traffic and 3 times the resolution. This is possible since only compressed motion videos are transfered over the network. By using the extra RK3066 core for compression you get about 10 times smaller files using XVID compared to MJPEG. As better video compression is implemented expect smaller files.
Finally Front Side camera is using 1280x1024 at 5 FPS. It averages only 37 Kbps for the same 24 hour period. If you only sent files that had just people detected it would use a lot less bandwidth. It really depends on the conditions and features you have enabled in CVP.
