Making Megapixel Cameras Shine in Low Light

Algorithms
Advances in the semiconductor manufacturing process, chip design, testing and packaging have enabled the industry to use ever-shrinking pixel sizes, thus allowing a dramatic increase in pixel count per camera. “Over the years, different techniques have been developed to try and deal with the degradation of image quality that results from shrinking the pixels,” Yanai said. Techniques include improved pixel sensitivity, binning several pixels together to create larger “virtual” ones and image enhancement algorithms. They all help, but cannot restore the image to its original quality.

While the image sensor determines the low-light performance, the processor decides how stable and clear images are. “Low-light processing takes place not on the image sensor, but on a separate DSP or SoC chipset,” Chang said. “CMOS chipsets require more loading compared to CCD ones, affecting processor efficiency.”

Network cameras have a great deal to do with software and firmware, compared to analog ones. Some megapixel models can maintain color rather than switch to black-andwhite imaging, for greater detail. However, increasing the video signal for nighttime performance also requires decreasing noise. “There are 2-D and 3-D digital noise reduction (DNR) algorithms, which produce different results,” Chang said. “The 2-D DNR is flat, while 3-D is a more complex and requires additional processing.”

Specialized chipsets can handle more complex processes. Some support night modes for optimization and software compensation, said Hee-Jun Lee, Executive Sales Manager, iCanTek.

The camera's algorithms determine whether video is high-quality. “Before compression, you have to optimize the ISP to preserve image quality,” said William Ku, Director of Brand Business for Vivotek. “The process defines the quality of the images, and then H.264 technology compresses and then restores images. Both technologies are important for the camera to capture the true images.”

By light sensitivity alone, analog and network cameras perform at about the same level nowadays, Ku said. However, the ability to provide megapixel images allows network cameras to provide much more detail.

Lenses, IR
High resolution puts unique demands on optics. Megapixel lenses must provide sufficiently high resolution to match the image sensor. Adding IR to the mix requires specialized — and pricey — IR megapixel lenses.

IR lenses need IR correction and coating. “A good IR lens should have the IR cut filter in/out controlled, which also makes image processing important,” said Akihito Yoshida, Sales Manager, Yamano Optical. This applies to both SD and megapixel models, with present SD sales outpacing megapixel. As resolution gets higher, costs for higher-quality materials must be managed carefully.

From the optics side, dome bubble refraction will also affect IR image quality. “The LED type and placement affect the camera's performance,” Ku said. “It's important to fit LEDs to the camera's angle, enabling the illumination to reach the maximum distance.”

Auxiliary IR illumination depends on how they are placed. IR arrays around the lens are cheaper to produce because of their compact size, but are affected by heat dissipation, Chang said. External LED lamps or “ears” placed outside of the lens' housing will cost more, but will prolong the camera's life span.

PoE
Most megapixel cameras run on PoE — the most common 802.af version delivers 15 watts. This is fine for 60 to 70 percent of surveillance applications using built-in IR LEDs for up to 30 meters. “For a powerful LED with an effective distance of 50 meters, it will require an external power supply,” Ku said.

Megapixel cameras have the low-light sensitivity to record decently. As they approach analog performance, the next article takes a look at whether lux ratings can be believed, as well as the outlook for low-light megapixel cameras.