The big squeeze
Although wireless video surveillance offers many benefits — such as the ability to monitor remote locations, quickly deploy cameras where needed or keep watch on an officer making a traffic stop — lack of bandwidth stands in the way of many implementations.
New spectrum available for public-safety broadband applications in the 4.9 GHz band, and in the 700 MHz band beginning in 2009, theoretically will make wireless video surveillance feasible. Still, that new spectrum is a finite resource, and video tends to use it up fast. Any agency considering wireless video surveillance should evaluate the techniques that different systems employ to optimize their use of the bandwidth.
One solution for transmitting video efficiently is to compress it, either by removing pixels from each image or removing frames from the sequence.
Nearly all surveillance video gets compressed to some extent, even the locally recorded video a merchant gives to police after a robbery, said Jonathan Hak, a crown prosecutor for the province of Alberta, Canada, and an expert on the legal aspects of forensic video in the U.S. and Canada. “If they’re getting a DVD from a store, the police agency needs to understand that it was already compressed by the store to begin with. And then it was compressed when it was put on the DVD.”
Euclid Discoveries announced a year ago that it has developed a video compression scheme that offers a 460% improvement over the MPEG-4 standard (MRT, March 2006), which would enable efficient wireless video transmission. For now, Euclid is targeting applications such as video conferencing and entertainment, but the technology could make it feasible to transmit surveillance videos over networks with limited bandwidth, the company said.
Public-safety agencies, however, should approach extreme compression schemes with caution. “High compression accommodates wireless transmission, but it severely reduces the quality of a recorded image if it is necessary as evidence,” said Mike Fergus, technical assistance coordinator for the Video Evidence Project of the International Association of Chiefs of Police (IACP).
To get a judge to admit a piece of video as evidence, a prosecutor must demonstrate that it’s both accurate and reliable, Hak said. That’s a problem when compression produces a low-resolution image. Even if it’s admitted, the jury often isn’t convinced.
“Someone robs a bank and is seen outside, but he just consists of two or three pixels in the shape of a block,” Hak said. “We could certainly say there was one person that left the bank, but we could never tell you anything about that person.”
Even a good image might not pass muster. “You could have a crystal clear video, and the defense could present an expert who could cause the court to be concerned about the reliability of the video,” said Grant Fredericks, a forensic video analyst with Forensic Video Solutions and digital media adviser to the IACP. For example, compression schemes that delete some frames use information in the surrounding frames to interpolate motion, which can create strange artifacts, such as a head that briefly separates from a person’s body, he said.
There is no specific standard to dictate which videos are admissible as evidence; each judge decides on a case-by-case basis, Fredericks said.
Compression is far from the only optimization strategy for wireless video transmission.
Boundless Security Systems, which sells wireless surveillance systems, records three versions of the stream from each camera its customers place in the field. These include a high-resolution “investigations stream,” a medium-resolution “monitoring stream” and a low-resolution “situation assessment” stream. The company offers systems that operate on cellular, satellite or Wi-Fi networks.
The strategy is to use no more wireless capacity than is necessary at the moment. “There is zero network traffic unless someone requests it,” said Steve Morton, the company’s CEO
For instance, if a police officer with a mobile surveillance camera in his car wants to call upon a second pair of eyes during a traffic stop, a dispatcher can use the wireless link to monitor the situation in real time. Typically, he or she would view the low-resolution stream, which, according to Boundless Security, needs 250 kb/s to transmit video at 30 frames per second (f/s).
If the situation merits a closer look, the dispatcher can view the video in greater detail by choosing one of the higher-resolution streams. “It will take some extra seconds to move that higher-quality, and thus higher-data rate, recorded stream,” but that doesn’t pose a problem, Morton said, adding that the system pulls the stream from the recording in the video server. A detective could study a piece of video during an investigation by using the wireless network to access part of the recorded high-resolution stream. A prosecutor also can present that high-quality recording in court.
About a dozen companies are working on technology that would automatically make the kinds of decisions that humans make in the example above, said Bill Lawrence, CEO of Armida Technologies, which offers wireless video surveillance systems on Wi-Fi networks.
Called video analytics, this technology puts intelligence in the camera controller to determine when activity in front of the lens offers cause for concern. “It’s not just about something moving. It’s about something very specific moving in a particular direction or crossing a threshold — or leaving a bag unattended for a period of seconds — and letting the system determine if that’s important, or not important — and notify someone,” Lawrence said.
Another strategy for managing wireless video transmission is to increase the network’s usable bandwidth. Armida uses a proprietary technology called Videlity to better manage the efficiency of a video transmission across a wireless network.
An IP video transmission at 30 f/s requires 1.5 Mb/s of bandwidth, Lawrence said. The typical 802.11g Wi-Fi network promises 54 Mb/s but actual throughput is much less. “Some of the reason for the inefficiency is the way that Wi-Fi typically manages the amount of total bandwidth available,” he said.
Videlity’s focus is on ensuring that all the data packets in a video stream make it across the network. “All wireless networks drop data packets which, if not recovered, will result in bad video at the other side,” Lawrence said. Recovery techniques built into wireless networks tend to produce video that appears jerky or pixilated, he said.
Armida adds proprietary error correction, protocol enhancements and other technology to improve upon the network’s own techniques, without increasing the time it takes to deliver the video across the network, Lawrence said.
Firetide also focuses on managing the wireless data flow. It develops fixed mesh networks to which users can connect surveillance cameras. Firetide offers a raw data rate of 70 Mb/s. Although throughput in a mesh network degrades as the signal hops from node to node, Firetide’s new HotPort 6000 dual-radio mesh nodes can be configured to maintain throughput at 35 Mb/s.
Not all wireless video surveillance systems deliver video as the action unfolds. L-3 Communications Mobile-Vision sells an in-car video recording system for the public-safety market. When a vehicle returns to a police station where the agency has installed a wireless access point, the system automatically uploads the recorded video to a server.
In this kind transmission, which takes just a few minutes to upload an hour’s worth of video, there’s no pressure to trade image quality for speed. “You’re always transferring an entire file, so the fact that you’re wireless does not affect the quality of the video itself,” said Chris Kodach, chief scientist at L-3 Communications Mobile-Vision.
For L-3, then, efforts to manage the bandwidth focus on managing upload traffic in the parking lot. If the network can’t accommodate video from all the onboard systems trying to transmit at once, it sets priorities based on what it knows about each officer’s activities. Consequently, the first car into the parking lot doesn’t always upload first, Kodach said. “Maybe there’s another one that’s got a higher priority because it has to turn around and go quickly. Or maybe there’s a high-priority incident on there that has to come off first.”
Ultimately, public-safety agencies will want to let someone sitting at headquarters monitor the view through an in-car camera lens in real time, and that’s gong to require a lot more bandwidth, Kodach said. To manage this kind of transmission, he envisions a hybrid strategy that would transmit at perhaps 5 f/s for remote situation assessment, but capture images in the field at 30 f/s.
“You’re recording it at the higher rate, because that will ultimately … be your evidence,” Kodach said.
All video streams are produced all the time for each camera (all parameters are programmable)
Investigations stream: Has sharpest images with highest resolution and fewest artifacts; 5 f/s (typical) and highest data rate (typically 0.5 Mb/s, maximum 2 Mb/s)
Monitoring stream: Has medium resolution, smoothest motion at high frame rate (10 f/s to 30 f/s) and medium data rate (typical 192 kb/s, maximum 750 kb/s at 30 f/s)
Situation assessment stream: Has lowest resolution and data rate (typical 32 kb/s, maximum 64 kb/s at 10 f/s, or 125 kb/s to 250 kb/s at 30 f/s)
Source: Boundless Security Systems