Q+A CENTER

It goes on to say that “the H.264 standard reduces the amount of information required to reproduce a video. Encoders process each frame subdividing the picture into a grid of blocks and searching previous or future frames for each block for matching texture, a technique known as motion estimation. When a suitable match is found, a decoder can reproduce the texture of the block in the current frame using only a vector pointing to the matching reference texture along with some information to correct any small texture differences. Where motion estimation fails to find suitable matches, encoders use the texture of nearby blocks in the same frame to predict the block texture and store the difference between the prediction and the actual texture. This is more efficient than storing the texture directly, but is more costly than motion estimation. The encoders act as ‘lossy’ compressors; their goal is not to reproduce the original picture exactly, but instead to choose the optimal means to reduce the data rate while preserving visual quality as best as possible.”

This kind of compression, says Damhaut, is common in visual surveillance, but not machine vision. “The reason is, you do measurements on images, but you don’t know what will happen after you compress the image. In surveillance, you can compress it without [worrying about] distortion of the image. Say you store the images of a parking garage overnight. If there’s a bit of distortion in the image, it’s not a problem.”

Damhaut says that Euresys, which is headquartered in Angleur, Belgium, recently completed a large installation in a Mexican city’s metro station. The system includes 4,500 analog cameras connected to a Euresys Picolo V16 board. “The signal is digitized, compressed and stored locally, and broadcast over a network to the central station,” Damhaut explains. “[The station] has huge video walls. They can select a camera and display the image. You need to store and retrieve an image of good quality. It has good bandwidth. The tradeoff will be how many days, weeks or months you want to store it, so the customer will choose the bit rate. They might say, ‘I want to keep very high quality 2 megapixels per second.’ That’s low for broadcast, but good for surveillance. You don’t see the degradation.

“In bandwidth,” he adds, “it all comes down to bit rate. It’s programmable. You can compress the image and reach 4-500 kb/second up to 2 mb/second bit rates.”

Euresys uses a chip from Stretch Inc. in Sunnyvale, Calif. “They provide hardware and software,” Damhaut says. “It’s the IP code we have implemented in our card to carry out compression. It’s highly programmable; we have customized cards for different applications.”