Capturing Crystal Clear Images With Megapixel Technology
Editor / Provider: a&s International | Updated: 4/12/2011 | Article type: Tech Corner
Megapixel surveillance is not a new concept — its applications and benefits are starkly clear. What has changed are smarter cameras, taking advantage of the added pixels and a better understanding of illumination in real life. In the first of a two-part report, A&S examines how smarter megapixel cameras are getting; the second part looks at best practices for optimal performance.
The big picture for megapixel surveillance cameras looks bright, in the wake of the recession. HD and megapixel cameras are expected to make up nearly 30 percent of network camera shipments in 2011, according to IMS Research. By 2015, it is forecast that more than 60 percent of network cameras shipped will be of megapixel resolution.
The resolution increase has a noted effect on the whole surveillance system. While a 2.1-megapixel or 1,080p HD image is six times larger than a D1 image, the additional pixels require a bigger pipe to transmit more data. The infrastructure and storage costs for megapixel are well-documented, with ROI and TCO being used as arguments in favor of bigger pictures. The fate of megapixel is linked to the future of IP networks, with HD forecast to make up most high-resolution cameras compared to megapixel, according to IMS.
Megapixel surveillance requires careful planning, but the benefits of added resolution boost the accuracy of analytics. Edge devices take advantage of faster processors, resulting in smarter use of pixels. Analytics can help reduce bandwidth, as an event will trigger video streaming, rather than constantly sending the same still images over the network. A more distributed architecture puts less strain on networks and makes life easier.
Clarity is the main driver for megapixel. “At the end of the day, you're putting in a security system to protect life and provide evidence in a court of law,” said Stephen Moody, Security Development Manager for ViS Security Solutions, an integrator in Ireland.
Cracking the Code
H.264 is the de facto standard compression for megapixel cameras, due to its efficiency in crunching large data files into smaller ones for transmission and storage. As compression evolved from M-JPEG's stills to MPEG-4 and now to H.264, a variety of profiles yield differences in performance. With 17 profiles in all, three are the most common: baseline, main and high, said Sachin Khanna, PM for CCTV, Bosch Security Systems.
By profile, the baseline is appropriate for video conferencing; the main profile is good for broadcast video; and high profile is most applicable for HD broadcast video. “H.264 requires a fair amount of processing power for encoding and decoding; this may limit the camera's frame rate and dictate the NVR platform to achieve the desired performance,” said Rich Pineau, CTO of Oncam Global.
Most H.264 profiles stem from 2-D applications, with not all profiles being capable of integration. “Even if both cameras are H.264 and the manufacturers are partners, the system could still not work,” said Patrick Lim, Director of Sales and Marketing for Ademco Far East. “The I/O and output are hard to integrate. Some engineers say it's easy to plug and play — there's no such thing.”
[NextPage]Different H.264 profiles are useful and important to maximize performance and efficiency, which affects camera selection. “Outside the top end of applications, people evaluate camera performance by visually comparing images from the cameras with different profile settings,” said Andrew Pigram, Technical Director at Norbain. “Ultimately, users want the highest quality for the lowest bandwidth, but will choose the right video stream profile for recording, transmission and live viewing.”
The compression's performance is marked by how well it captures moving images, without spiking in bit rate. “There are a variety of simple yet effective ways to evaluate encoder performance, such as waving a hand in front of the camera or panning the camera itself, like simulating a speed dome,” said Bengt Christensson, Senior Director of Marketing, Ambarella.
Baseline and high profile H.264 meet real-time requirements of about 100 to 200 milliseconds, with longer delays for megapixel, said Xiang Wei, CTO and VP of Engineering, Grandstream Networks.
Using H.264 permits megapixel cameras to match standard definition (SD) bit rates. An SD PTZ camera using MPEG-4 compression would need about 2.5 to 4.0 megabits per second (Mbps); the bit rate would fall to 1 to 1.75 Mbps using H.264, keeping storage manageable. “Most cannot see a difference between 1.5 Mbps and 6 Mbps in H.264 in side-by-side testing,” said Ed Thompson, CTO of DVTel.
However, some integrators prefer to work with MPEG-4 over H.264, minimizing any effects on quality. “It depends on the surveillance and recording platform because one of the serious cons is not every manufacturer supports the H.264 codec, particularly in remote monitoring situations,” Moody said.
No single compression profile will make or break a camera — it boils down to the viewing purpose. Nearly all megapixel cameras support multiple streams, allowing users to choose the appropriate one for different situations. “During off hours, a building could be monitored with M-JPEG at a slow frame rate, and switch to H.264 at a faster frame rate when an alarm is triggered,” said Becky Zhou, APAC Sales Director for Arecont Vision.
H.264 compression is less efficient with motion and also loses detail, while M-JPEG does not lose a frame. “If you have a lot of movement and every pixel changes every second, H.264 predicts what happens and what takes place in the scenery,” said Gerrit Schreiber, Senior PM at Basler. “But with M-JPEG, you are looking at the detail. For a plain surface, M-JPEG is most clear. But if you are looking at a mosaic with a high amount of detail, M-JPEG will increase bandwidth requirements.”
“M-JPEG is somewhat like the cockroach of the industry: It isn't very pretty, but it's very good at what it does, and it isn't going anywhere,” said Ian Johnston, CTO of IQinVision.
The lifespan of M-JPEG will be extended by increased storage. “There's going to come a time when cloud storage becomes so cheap and everyday, storage constraints will be a thing of the past,” said Todd Pinnell, PM for Video at Speco Technologies. “It's what you want to see and what you want to pay.”
One of the most noticeable trends in megapixel surveillance is intelligence. Analytics are being baked into the latest DSPs from Texas Instruments and Hisilicon, making basic algorithms — people counting, motion detection, missing object detection and tracking — nearly universal on megapixel cameras. “Basic intelligence would be motion detection; then, next are some more advanced analytics that go beyond to things we haven't thought of yet,” said Vance Kozik, PM at StarDot Technologies.
Analytics take advantage of faster processors and increased pixel counts. This makes edge devices suited for detail recognition, particularly for facial or license plate recognition that requires zoom, said Philip Siow, Senior Consultant for South APAC, Axis Communications.
While most megapixel cameras support basic analytics, they will support more intelligent access control and intrusion functions, said Ebony Huang, President and CEO of Brickcom.
HD or megapixel cameras are expected to become smarter with analytics and storage on the edge, changing video system architecture from server-centric to edge-centric and providing reductions in TCO. “With fewer servers to buy, install and maintain, coupled with a lighterweight network when not streaming the recording video back to the servers, the system becomes cheaper and easier to install and maintain,” Thompson said.
A smarter camera that puts less stress on the network will enable megapixel cameras to handle 24/7 recording. “There are also a few trade-offs to consider,” Christensson said. “Higher resolution and pixel rates require more processing, and features such as WDR will further add to the DSP cycles needed, requiring low-power ISP implementations.”
Higher pixel density will require higher processing power for the infrastructure, said William Ku, Director of Brand Business for Vivotek.
Objectively, there are some benchmarks for identifying a megapixel camera with a good sensor. “Look at the pixel size,” Schreiber said. “Do not trust any of the sensitivity figures, because there is no standard.”
To combat poor low-light sensitivity, IR illuminators boost image sensor performance. “For side-by-side low-light performance, a CCD is more light-sensitive than a CMOS sensor,” Kozik said.“However, that is beginning to change.”
More pixels in megapixel sensors will generate more noise at night. “The performance with IR lighting for night conditions is quite important because there are some megapixel implications, particularly with IR-cut filter switching sometimes not being as smooth as a typical CCD camera,” Moody said.
Megapixel cameras are constrained by the optical limits of lenses. True megapixel lenses are rare, expensive and therefore have not kept pace with image sensor increases. “One thing people use megapixel cameras for is the wider view requiring fewer cameras, but it's a challenge to do that with more background motion,” Moody said. “Lens adjustment in external areas of a large warehouse or retail situation may pick up too much background motion interference, with no relevance for the site's consideration. It is important that the scene does not overspill onto a surrounding road, where there may be excessive traffic flow.”
There are 5- or 10-megapixel lenses for machine vision, but they can retail from US$700 to $1,400 each — about the same price or even more than the camera itself. “There's not a standard lens that can provide the resolution for a 5-megapixel image sensor,” Schreiber said. “The pixel size is so small at 2 microns that it cannot be held with a standard lens. I can give you an example of a 5-megapixel camera using a bad lens resulting in a worse image than a 1.3-megapixel camera with a good lens.”
A good megapixel lens should have the same resolution at the edges as well as the center, Khanna said. Image performance should be maintained throughout lighting changes, particular for outdoor scenes that require IR correction.
Fisheye lenses are catching on in megapixel surveillance, thanks to increased sensor sizes. A good hemispheric camera should select a fisheye lens that is responsive and matches its sensor, Pineau said.
Looking at the spec sheets is good reference, but nothing beats field tests. “We do quite a lot of large projects, where the client requests 100 to 200 cameras, so this type of application warrants looking at megapixel or HD cameras to reduce the camera quantities,” Moody said. “We go out with footage, show them how it was tested in various conditions and show them the results for cameras 1, 2, 3 and 4. We will show them the one we would recommend; it is important that the client understands the choice and why we have recommended a certain type of camera over another.”
On paper, there are relatively few differences between cameras. However, software and programming will affect performance; two cameras using the same hardware can produce completely different results. In our next section, we explore best practices in megapixel camera installation, particularly for illumination, as well as real-life project specifications.