Sanyo Reveals Secrets of World's Smallest, Lightest Camcorder
Sanyo Electric Co Ltd announced the miniaturization and weight saving technologies used for "DMX-HD1000," the latest model of its "Xacti" series camcorders, at the press conference Dec 6, 2007.
The DMX-HD1000 is "the world's smallest and lightest" full HD (1920 × 1080i) camcorder, the company said. Its volume is 272cc, and its mass is 268g. It was released in September 2007.
The miniaturization and weight saving could be achieved because the image processing functions, which are usually allotted to several chips, were integrated in one chip. This reduced not only the volume and mass of the LSI but also the number of heat radiators for the LSI.
The DMX-HD1000 is equipped with "Platinum Σ Engine," a chip that can rapidly process movies (MPEG) and pictures (JPEG) in parallel by sharing the memory bus.
One of the major factors that contributed to the integration of the chips is a self-developed H.264 (MPEG-4 AVC) codec that enables low power consumption, small-sized circuit and high compression. This codec reduced the power consumption of the full HD camcorder by one-third and the circuit size by 40%, and enhanced the data compression ratio of MPEG-4 by 50% to about 1/100.
With a 4-Gbyte SD card, 45-60 minutes of movies can be recorded.
Three major technologies were used to develop this codec, (1) a technology to increase the efficiency of motion vector detection (MVD), (2) an architecture for parallel motion detection and (3) an algorithm to control optimum code amount.
As for (1), different detection methods are used in slow-moving scenes and in fast-moving scenes to reduce the calculation amount. In slow-moving scenes, the surrounding area of the point where the object was originally located is searched. In fast-moving scenes, the direction and amount of panning are estimated to decide where to search.
A self-developed technology is also used for (2). Conventionally, to search for motion vectors, each macroblock (16 × 16 pixels) is processed, and the data are transferred by one circuit. In the new method, two adjacent macroblocks are processed at the same time. And two circuits are used to transfer data to reduce the data traffic (bandwidth) by 50% (30% for H.264 data as a whole).
(3) was realized by conducting a three-dimensional (lengthwise direction, crosswise direction and time) correlation analysis of pictures. When there are fast-moving parts and slow-moving parts in a picture, usually, more code amount is spent for the former. However, when an unimportant "background" is moving fast and an important "human face" is moving slowly, the face blurs regardless of its importance.
Slow-moving areas without a patchy pattern are more likely to be considered as important, Sanyo Electric said. The correlation analysis detects those areas, and more code amount is spent for them. Therefore, more satisfactory picture quality is achieved by using the same amount of code.
Also, electronic compensation, not optical compensation, was adopted to avoid blurring of images due to hand movement. There are some electronic compensation methods such as "image restoration method," which forecasts a static condition from the angular information collected by a gyro sensor, and "picture addition method," which takes some pictures and compound them.
The picture addition method, which does not require a gyro sensor, was adopted for the DMX-HD1000 to reduce the size and weight of the camcorder.
Correction Notice: In the original article, we incorrectly stated that "the image restoration method" was adopted for the DMX-HD1000. In fact, it is not the image restoration method but "the picture addition method." We apologize for the mistake and regret any misunderstanding it may have caused.